Unusual Whales, citing Bloomberg, reported on November 13 that banned Nvidia chips ended up in a Shanghai AI data center linked to Alice Huang. The post says a company she founded, now majority owned by her sister, listed 52 Super Micro units with H100 or H200 chips, and that a Singapore Nvidia cloud partner bought 22 Dell servers with H800 chips that went to a Shanghai facility run by her husband, Kang Wei. This is a separate report from the Reuters review of more than $3B in Chinese projects built around restricted Nvidia chips, which is also in this deck.

Unusual Whales posted on November 13, citing Bloomberg, that banned Nvidia chips ended up in a Shanghai AI data center linked to Alice Huang. According to that post, a company Huang founded, now majority owned by her sister, listed 52 Super Micro units with Nvidia H100 or H200 chips, which cannot be sold to China without US approval. The post also says a Singapore Nvidia cloud partner bought 22 Dell servers with Nvidia H800 chips, and that the chips went to a Shanghai facility run by Huang's husband, Kang Wei.

Those details are secondhand: they come from Unusual Whales' summary of Bloomberg, not from a post by Bloomberg itself or from a document published in the scanned posts. The post does not name the Singapore partner or the Chinese company, and it does not say whether Alice Huang personally arranged the purchases.

Unusual Whales also wrote that US officials suspect Alice Huang was recruited by Chinese intelligence to target Jensen Huang, who resigned as Nvidia CEO this week. That allegation is unconfirmed in the posts scanned and should be treated as a claim reported by Unusual Whales, citing Bloomberg, not as a proven fact. This report is a separate event from the Reuters review, also in this deck, of more than $3 billion in Chinese computing projects built around restricted Nvidia processors.

Watch for Bloomberg's full article, for named responses from Super Micro, Dell, Nvidia and the Singapore partner, and for any US or Chinese official comment. Documents that identify the companies, the facility and how the servers moved would make the account easier to check. Congressional testimony by Jensen Huang, already expected in other coverage this week, may also take up the export control questions raised here.

Nvidia's official account posted on November 13 introducing GR00T 2, which it called its most advanced, open System One model for robotics research. The post says the model learns skills from human demonstrations, predicts how its actions will change the physical world, and can carry those skills across different robot bodies. Nvidia published a research page with its own test results for the model.

Nvidia's official account posted on November 13 introducing GR00T 2 as its most advanced, open System One model for robotics research. The post says the model learns skills from human demonstrations and predicts how its actions will change the physical world, and that those skills can carry across different robot bodies, reducing the need to teach each robot every task from scratch.

Nvidia's research page describes GR00T 2 as a foundation world action model that predicts future video and robot actions together, and says the models run on parallel grippers, dexterous hands and humanoids. According to that page, on 50 tasks never seen on the target robot, GR00T 2 succeeds 40% of the time with no demonstrations on that robot. Five demonstrations per task, which Nvidia says is under 3 minutes of robot data, raise that figure to 95% after a 30 minute fine tune on a single RTX 5090.

Nvidia also describes a faster variant, GR00T 2 Turbo, that it says can run on a Unitree G1 humanoid, with each prediction taking 124 ms on the robot's own Jetson Thor. Those figures come from Nvidia and have not been independently checked in the posts scanned.

Watch for Nvidia to publish weights, code or training recipes, and for other labs to try the model on their own robots. Independent tests of the official zero shot and few shot figures, and of skill transfer across robot bodies, would show whether the results hold outside Nvidia's setup. Follow up posts in the official thread may add demos or documentation links.

The official ChatGPT account posted on November 13 that a user's dot can complete purchases on their behalf once the user approves, and that ChatGPT Wallet lets agents pay without access to actual card details. OpenAI's accompanying announcement says the wallet was built with American Express, Mastercard and Visa, and that it is rolling out first to ChatGPT Pro subscribers in the US. Every purchase, OpenAI says, requires a security check such as Face ID or a fingerprint scan.

The official ChatGPT account said on November 13 that a user's dot can complete purchases on their behalf with the user's approval. The post says ChatGPT Wallet lets agents pay without access to actual card details, that every purchase requires a security check, and that the wallet is rolling out starting that day to dots, for Pro subscribers in the US.

OpenAI's announcement adds that the wallet was built in partnership with American Express, Mastercard and Visa. According to OpenAI, a user adds a card once, then reviews the items, the total and the selected card before each purchase and approves it with a security check such as Face ID or a fingerprint scan. After approval, the payment network provides a payment token, a substitute for the user's actual card details, and the charge goes to the selected card without the wallet sharing those details with the business or the agent. OpenAI says users keep existing card rewards and protections for eligible disputes, and that users transact directly with businesses.

OpenAI says the wallet will initially be available to ChatGPT Pro subscribers aged 18 and older in the US with eligible US issued cards, and that it is starting with limited availability so it can learn from real world use. The company also published a Commerce Partner Playbook for businesses and says it is working with fraud prevention and payment partners so risk systems recognize legitimate purchases made by agents.

Watch whether OpenAI expands ChatGPT Wallet beyond Pro subscribers in the US, and whether major retailers and payment systems accept the tokens without declining the charges. Official comments from American Express, Mastercard and Visa already appear in the announcement; later reports from banks or merchants will show how often purchases go through. If OpenAI later lets users set spending limits or categories in advance, that would be a further change to report here.

DeepSeek already offered a consumer App and an API for its models. On November 13, the official DeepSeek account said the capabilities behind the App, including search, files, links, safety and voice, are now available through the DeepSeek API. The company also posted an open source frontend demo and a guide to building a chatbot with those features.

DeepSeek posted on November 13 that the full DeepSeek App experience is now available through its API. The lead post lists search, files, links, safety and voice, and says developers can enable built in features with a few parameters or call standalone APIs such as OCR, ASR and TTS, including to recreate the full App in their own application. The post also links to an open source frontend demo that can import DeepSeek conversation history, to the demo's GitHub repo, and to DeepSeek's Build a Chatbot guide.

In a follow up post, DeepSeek said V4.1 Flash, the model listed as deepseek-flash, now supports these capabilities across the Messages, Responses and Chat Completions APIs, with tool calls and context assembly handled on the server. The post lists web search (web or images, with citations, when thinking is off), deep thinking search (search and read pages through multi step reasoning), file input (PDFs, Office files, images and more), URL reading (fetching links in user messages) and safety checks that screen inputs and outputs and return refusal indicators when risks are found. DeepSeek said public facing apps must enable safety checks, and that developers remain responsible for content moderation, applicable model registration and AI generated content labeling. Search, Fetch, OCR, ASR and TTS are also offered as standalone APIs.

A third post in the thread said extended capabilities incur tool usage fees on top of standard token charges. The pricing image DeepSeek shared lists Search and Fetch at $0.75 per 1,000 requests, OCR at $0.225 per 1,000 pages, ASR at $0.058 per hour (marked 50% off for a limited time, down from $0.116 per hour) and TTS at $0.225 per 10,000 characters (also marked 50% off for a limited time, down from $0.45 per 10,000 characters).

DeepSeek's API docs describe the same feature set and say the service runs the searches and URL reading, while the client shows the returned blocks and passes them back unchanged. The docs also say the open source demo stores the API key, conversations and uploaded images only in the browser.

Watch for developers to ship apps that reuse the DeepSeek App's search, file and voice features, and for whether the standalone OCR, ASR and TTS APIs get used outside chat products. The limited time ASR and TTS discounts may end, which would change the cost of voice features. It is also worth watching whether DeepSeek extends the same built in tools to models beyond V4.1 Flash, and how safety check requirements are enforced for public apps.

Reuters reported on November 13, citing an internal Amazon document and people familiar with the matter, that Amazon is planning an initiative called World Product Graph. Watcher Guru shared the news on X. AI shopping agents from rivals such as Google and OpenAI are starting to compete for where people research and buy products.

Watcher Guru posted on November 13 that Amazon plans to collect data on nearly every product sold on earth to compete with Google and AI shopping tools. The post is based on a Reuters exclusive published the same day.

Reuters reported that the initiative, called World Product Graph, seeks to gather data on virtually every item for sale, whether Amazon stocks it or not, according to an internal document and people familiar with the matter. The document, written like a press release, carries an aspirational release date of October 1, 2027, and Reuters said it could not determine if or when the plan will be carried out.

According to Reuters, the plan relies on brands and retailers contributing product data and on Amazon crawling other retailers' websites. The document says Coca Cola and Bayer have signed up to contribute data. It also says many answers from Amazon's AI shopping assistant Rufus are factually wrong or incomplete, and that running its roughly 83 million weekly searches is expensive. Amazon told Reuters the usage figures are significantly outdated and that its data shows Alexa for Shopping is about 97% accurate globally.

Reuters cited a 2025 Morgan Stanley report showing 16% of online shoppers start product research on Amazon compared with 57% on Google.

Watch for Amazon to confirm the plan or launch parts of it, and for how brands and retailers react to Amazon crawling their sites. Legal questions over scraping and data access may follow. The plan also signals tougher competition with Google, OpenAI and Meta over AI shopping agents, an area where Amazon has already moved to limit outside agents on its store.

Natasha Bertrand of CNN reported on November 13 that US intelligence intercepts in the spring indicated an affair between Jensen Huang, then Nvidia's CEO, and an executive at Megaspeed, a firm under investigation for possibly diverting sensitive technology to China. According to sources familiar with the matter who spoke to CNN, officials asked for the American in the intercepts to be unmasked, the person was identified as Huang, and Secretary of State and National Security Adviser Marco Rubio drafted a classified memo that went to a small group of senior officials, including at the White House. CNN names the Megaspeed executive as Alice Huang.

Natasha Bertrand, a CNN national security correspondent, posted on November 13 and linked to a CNN report that US intelligence intercepts in the spring indicated an executive of Megaspeed was having an affair with a US person. Sources familiar with the matter told CNN that the American was first masked in the reporting, a standard privacy step, and that officials then asked for the name to be unmasked, revealing former Nvidia CEO Jensen Huang.

According to those sources, Secretary of State and National Security Adviser Marco Rubio drafted a memo summarizing the intelligence and sent it to a very small group of senior US officials, including at the White House. CNN says it is unusual for a US person's identity to be unmasked, and that such a request typically has to come from a very senior official.

CNN names the Megaspeed executive as Alice Huang and says Rubio wanted officials to be aware of the relationship when they dealt with Jensen Huang. Sources told CNN the memo was not written specifically for President Donald Trump, but that he was briefed more than once, including verbally by Rubio aboard Air Force One in May, when Huang joined the president's travel group to China.

A person familiar with Trump's response told CNN he appeared at least partly focused on the affair as an embarrassing personal matter and also voiced concern for the US economy. After Huang resigned, CNN reports, Trump said the former CEO was "going through hell," was concerned for his wife and children, and said the question of whether Alice Huang was a Chinese spy was unsettled.

CNN says Alice Huang had been an important National Security Agency target by the spring, and that Justice and Commerce began an investigation into her firm in 2024 as part of a broader probe into shell companies suspected of diverting restricted Nvidia chips to China. Officials still disagree on her relationship with the Chinese government, CNN has reported, and Attorney General Todd Blanche said on Thursday that Justice and the FBI have reached no conclusions about who she is or is not.

Watch for any official confirmation of the memo, and for what investigators say about Megaspeed and Alice Huang as Commerce and Justice continue their work. Nvidia reports third quarter results on November 17, and further official comment from the White House or from Huang could change how this report is read. Any public statement from Rubio, or a denial from Huang, would also test the sources who spoke to CNN.

Maggie Haberman of The New York Times posted on November 13 a passage from a Times story she wrote with Julian E. Barnes and Jonathan Swan. The passage says US intelligence concerns that Alice Huang was spying for Beijing were taken so seriously that President Donald Trump was briefed twice, late last year and again this spring. This is a separate story from David Sacks's later dispute with the Times. No official New York Times account post of this article cleared 100,000 views in this scan, so Haberman leads.

Maggie Haberman posted at 8:37 PM ET on November 13 and quoted a New York Times story published the same day, written with Julian E. Barnes and Jonathan Swan. The quoted passage says, "The concerns in the U.S. intelligence community that Ms. Huang was spying on behalf of Beijing were taken so seriously that Mr. Trump was twice briefed on it, late last year and again this spring." Haberman tagged both co authors and added "me," marking the post as coming from the reporting team.

The linked Times story is titled "Nvidia Scandal Prompts Questions of What It Means to Be a Spy." The card text on Haberman's post says US intelligence agencies believe a woman who had an affair with Jensen Huang was a Chinese asset, and that they are still assessing more about her role. Those are the Times's characterizations, not an official US finding in the posts reviewed here.

The official New York Times account did not post this article with 100,000 or more views in this scan. Julian Barnes reposted Haberman's post, but that repost had only 127 views. Haberman therefore leads, and the Times article is linked and attributed.

This reporting sits next to, but is not the same as, two other stories already in this deck. CNN, in a separate report led by Natasha Bertrand, said President Trump was briefed more than once, including verbally by Secretary of State and National Security Adviser Marco Rubio aboard Air Force One in May. The Times account adds a count of two briefings and an earlier window, late last year, as well as this spring. David Sacks later argued that Times coverage of Alice Huang relied on unconfirmed claims; that dispute is a separate opinion story.

Jensen Huang resigned as Nvidia's CEO on November 12. Officials, including Attorney General Todd Blanche, have said in earlier coverage that Justice and the FBI have reached no public conclusion about who Alice Huang is or is not.

Watch for any on the record confirmation from the White House or from US intelligence that those two briefings took place, and for how the Times timeline sits beside CNN's account of a May briefing on Air Force One. Nvidia reports third quarter results on November 17, which may bring questions about what the board was told. A public statement from the administration, or a denial, would test the Times's unnamed official sources.

Megaspeed International is a Singapore based cloud company that US agencies suspect of routing restricted Nvidia chips to China, and CBS News reports it was previously led by Alice Huang, whose relationship with Jensen Huang preceded his resignation as Nvidia CEO. The Commerce Department controls the US Entity List, which restricts access to American technology. Megaspeed was still not on that list as of November 13, according to TokenPost.

Treasury Secretary Scott Bessent told CNBC's Squawk Box on November 13 that the Treasury, War and State departments and the Office of the US Trade Representative "were in favor of putting Megaspeed on the entity list for months." According to CBS News, he added that "Commerce has the pen" and that questions about its internal processes should go to the Commerce Department, which Howard Lutnick leads.

CNBC reported that the remarks followed an Axios report that Bessent had asked Lutnick in the White House Situation Room why he resisted export controls on Megaspeed. A Commerce spokeswoman told CNBC there was "no such conversation," and Bessent said, "We don't repeat conversations in the Situation Room." He said he was frustrated because President Trump had made clear that China should not receive Nvidia chips above H100s, and that any leakage hurts US companies in the AI race.

Bessent also turned to Nvidia itself. "The real story here should be: Did Nvidia know their client," he told CNBC, adding that anyone at the company should be held accountable if it sold to obvious shell companies acting as pass throughs. TokenPost reports that Megaspeed denies illegal transfers.

CBS News reported, citing a US official, that Lutnick agreed to testify before the Senate Banking Committee on December 10. CNBC said Sen. Tim Scott had announced that both Huang and Lutnick were asked to testify.

Watch whether Commerce adds Megaspeed to the Entity List before Lutnick's Senate Banking Committee appearance on December 10, and whether the White House settles the dispute in public. Nvidia is set to report third quarter results on November 17, its first report since Huang left, where customer screening and China exposure could draw questions. The Justice Department's criminal investigation of Megaspeed, reported by TokenPost, is a separate track that could move on its own timeline.

Georgi Gerganov, the creator of llama.cpp, posted on November 13 that llama.cpp v1.0.0 is now available, and linked to a Hugging Face blog post by the ggml org team. The blog says the 1.0.0 release covers both the llama.cpp inference library and the ggml tensor library, and that future major version bumps will mark incompatible changes to the public C API. The project has been used for local and private model inference for about three and a half years.

Georgi Gerganov posted on November 13 that llama.cpp v1.0.0 is now available, and linked to a Hugging Face blog post titled llama.cpp v1.0.0 : A stronger foundation for the local AI ecosystem. The blog, published the same day by the ggml org team, says the release highlights the project's responsibility to downstream developers who depend on its library interfaces.

According to the post, the 1.0.0 milestone applies to both llama.cpp, the C and C++ library for LLM inference (libllama), and ggml, the tensor library (libggml). Going forward, the team says both projects will use semantic versioning: a major version bump for a breaking change to the public C API in the include headers, a minor bump for backward compatible features or API additions, and a patch bump for bug fixes with no API change.

The blog says the project began about three and a half years ago and that llama.cpp currently supports more than 150 architecture variants as well as mainstream speculative decoding methods. It also lists distribution channels including Docker, Debian, Homebrew, conda forge, MacPorts, Winget and an XCFramework for iOS and macOS, and says hardware support for CI runners comes from NVIDIA, AMD, Qualcomm, Arm, IBM and Intel.

Watch the project's release notes and packaging channels for how strictly the new versioning is applied, and whether downstream apps start pinning to the 1.0 API. Follow up posts may cover desktop apps or additional backends. Independent packagers will show how quickly v1.0.0 lands in Homebrew, Debian, Docker and other channels named in the blog. If a later release has to bump the major version, that will test the new contract.

OpenAI had a robotics team in its early years, wound it down in 2020, and rebuilt one in 2024. On November 13, Fortune reported in an exclusive that OpenAI has disbanded that stand alone team, and the company confirmed the reorganization. The lead post here is from prinz, an AI commentator, sharing an excerpt; OpenAI itself had not posted about it on X in the posts scanned.

According to Fortune, OpenAI moved its robotics research team into its main research organization, reporting to Chief Research Officer Mark Chen, while the applied robotics and hardware teams moved to the Applied organization under President Greg Brockman. A company spokesperson told Fortune that robotics remains a core part of OpenAI's long term vision and described the change as a sign of commitment, not a retreat.

OpenAI told Fortune that at the start of the year its researchers believed the best path to robotics ran through models like Sora, but that its general frontier models, including GPT 6 Astra, have since shown they can be applied directly to robotics. Fortune cites an Nvidia study in which an Astra powered agent reached a 74.8% success rate on a simulated manipulation benchmark, compared with 51.0% for the best specialized robot models on that leaderboard.

Fortune reports that OpenAI had already registered booth 208 at CoRL, a top academic robotics conference in Austin this week, and that the booth sat empty when its reporter visited. The outlet later updated the story to clarify that former team members were present. It also notes that Sam Altman has targeted a first public demonstration of an OpenAI humanoid robot in 2027, and that the company is under pressure to cut costs, with reported operating losses of $10.9 billion in the most recent quarter.

Carnegie Mellon professor Guanya Shi told Fortune the move signals that OpenAI sees robotics as no longer a special problem. MIT professor Russ Tedrake said robotics companies still hold an edge in real world deployment data.

Watch whether OpenAI still shows its own humanoid robot in 2027, as Sam Altman has targeted, and how robotics now fits under Mark Chen and Greg Brockman. Results from general models running real robots, not only simulations, will test the bet. Robotics startups that raised money on the idea that robots need their own brains may face harder questions from investors.

A CEOInterviews.AI clip posted on November 13 shows SemiAnalysis founder Dylan Patel saying Jensen Huang personally persuaded even small suppliers to let Nvidia buy up their output, and that with Huang gone, Google is already asking those suppliers for more capacity. Elon Musk quoted the clip on November 14 and wrote that many of those deals were a handshake with Jensen, not a written contract, and that without him at least some of those critical deals will be lost. In a later post, Musk said that when Jensen gave his word on compute delivery or on speeding up components, he always came through.

CEOInterviews.AI posted a video on November 13 in which Dylan Patel, the founder of SemiAnalysis, describes how Jensen Huang handled Nvidia's supply chain. In the clip, as quoted in the post, Patel says Huang "had complete foresight and understood the demand required and went to all of these executives, even of tiny companies where they only generate a few hundred million dollars of revenue but they have an important subcomponent." He says Huang convinced them to let Nvidia buy the supply and build it out, and that with Huang gone those companies are looking around the market because "Google's begging for more capacity from the whole supply chain." Patel also asks whether Nvidia can keep its technology lead without Huang, saying the company is only about a year ahead of its biggest competitors.

Elon Musk quoted that post on November 14 and wrote, "Jensen drove supply chain brilliantly at Nvidia, but many of those deals were a handshake with Jensen, not a written contract." He added that without Jensen, at least some of those critical deals will be lost. In a reply a few minutes later, Musk said that if Jensen gave him his word about compute delivery, or if Musk asked for components to be accelerated, Huang always came through, and that a purchase order would be put together later.

These are Patel's and Musk's accounts of how the deals worked. Neither post names the suppliers, quotes a company confirming that a handshake is now in doubt, or shows that Google has already won capacity that Nvidia expected to keep. The comments land in the same week as other stories in this deck on Huang's resignation, the employee petition to bring him back, and US official concern about Nvidia's supply relationships.

Watch for any supplier, Google, or Nvidia comment that confirms or disputes the handshake account, and for signs of capacity shifting at earnings on November 17. If buyers start posting delays, or if Google or others announce new supply, that would be the first public test of Musk's warning. Until a named supplier speaks, the claim rests on Patel's interview and on Musk's first hand recollection.

On October 29, Cursor said it was investigating a security incident and a malicious, unauthorized copy of Anthropic's Claude Code extension in the Cursor VS Code extension marketplace. On November 13 it published a fuller account of the attack and its remediation. Cursor says it believes the incident is contained and that the last attacker activity it saw, on November 1, was blocked.

Cursor says the attack started when an attacker obtained an access key Cursor had issued to Fireworks, a training and inference vendor. According to Cursor, the key was exposed by Fireworks and used to copy data from a read only storage bucket shared with the vendor. The bucket held limited coding agent traces, mostly from 2025, from sessions not in Privacy Mode that Cursor was permitted to use for model training.

Some traces contained credentials that had not been redacted. Cursor says less than 0.01% of customers had API credentials exposed, and that agent traces were exposed for fewer than 0.05% of users not using Privacy Mode. With those credentials, the attacker reached select SaaS tools and from there accessed archived source code repositories not used in production, limited support data and emails, usage metrics and contact details, summaries of some agent conversations, and some credentials users had sent to support.

On October 29 the attacker used a stolen credential to place tampered copies of the Claude Code extension, versions 2.1.312 and 2.1.313, on the cache serving Cursor's extension marketplace. Cursor says at most 5,000 users may have received the malicious build, which was available for about 11 hours. Anthropic did not publish it, Cursor says it was not a flaw in Anthropic's product, and the Open VSX source never had the malware.

Cursor says production systems, the Cursor app and its official updates, customer repositories and live inference were not affected. It moved all inference off Fireworks, rotated credentials across internal systems, rebuilt the marketplace, added secret scanning and brought in outside forensics experts.

Watch for a response from Fireworks on how the key was exposed, since Cursor's account places the origin there and no Fireworks statement appeared in the posts scanned. Affected teams have been told which keys to rotate, and anyone who installed the tampered extension versions is asked to follow Cursor's security notice, because uninstalling alone may not be enough. Further updates to Cursor's post, regulatory notices or customer claims would mark the next stage of the incident.

Josh Rogin posted on November 13 quoting a New York Times report that Cantor Fitzgerald has served as a financing or placement agent for several companies involved in developing or operating data centers in Southeast Asia, which the Times said could be hurt by tougher US limits on chip sales to the region. Rogin added that the firm, run by the sons of Commerce Secretary Howard Lutnick, had a financial interest in looser chip limits that Lutnick was in charge of implementing. Official Times account posts on related Lutnick coverage in this scan were far under 100,000 views, so Rogin's post leads and the Times report is linked and attributed.

Josh Rogin posted at 9:56 PM ET on November 13 and quoted a New York Times report dated November 12. In the passage he highlights, "Cantor Fitzgerald has served as a financing or placement agent for several companies that are involved in developing or operating data centers in Southeast Asia… which could be hurt by tougher U.S. limits on chip sales to the region."

Rogin then offered his own reading: that the firm run by Lutnick's sons has a financial interest in looser chip restrictions that Lutnick himself was in charge of implementing, which Rogin called the clearest conflict of interest he has ever seen. The conflict claim is Rogin's. The underlying description of Cantor's role is attributed to the Times.

The linked Times article is at a November 12 URL on Lutnick, Nvidia, Commerce and China chips. This scan did not find an official New York Times account post of that article with 100,000 or more views. Related official Times reposts on Lutnick coverage had under 21,000 views, so they cannot lead.

The report lands in a week when Treasury Secretary Scott Bessent has said Commerce "has the pen" on whether to list Megaspeed, and when lawmakers have asked Lutnick to appear before the Senate Banking Committee on December 10. Those points are already in this deck from Bessent's CNBC interview. The Times passage, as quoted, does not name the Southeast Asian data center companies in Rogin's post, and this story does not add names that were not in the posts reviewed here.

Watch for comment from the Commerce Department, from Cantor Fitzgerald, or from Lutnick on the Times account of Cantor's work with data center firms in the region. Lutnick's planned Senate Banking Committee appearance on December 10 is a likely moment for questions. Any later Times detail that names the companies, or a filing that shows the placement work, would make the reported conflict easier to test. A decision on listing Megaspeed would also show how Commerce uses the authority Bessent said it holds.

The Wall Street Journal published a report on the evening of November 12, Eastern time, headlined "Before His Ouster, Jensen Huang Belittled the Executive Replacing Him." It describes Nvidia's financing deals and Huang's view of Colette Kress, who became interim CEO when he resigned. Bull Theory's viral post on November 13 summarized the report and paired it with a second clipping on Wall Street's AI financing plans.

According to the Journal, Huang summoned Kress and his senior finance team this summer to diagnose a growing problem: Nvidia was in talks to provide billions of dollars of guarantees to companies that could not afford its chips. The paper says that included deals led by Huang's son in law, Nico Caprez, who had been offering upstart cloud providers and data center developers financing in the hope of turning them into long term customers.

The Journal reports that in the spring, Caprez and vice president Raj Mirpuri told prospective partners that Nvidia could arrange credit support deals within a couple of weeks, that dozens of companies lined up, and that Nvidia proposed cloud providers share a cut of their revenue in exchange. Kress raised questions about which deals to pursue, how to value the guarantees and how to calculate returns, a person familiar with the matter told the paper.

During an August meeting, the Journal says, Huang called Kress "a deer in the headlights" and a "reporting CFO," not someone who could help define strategy.

Bull Theory's post also showed a clipping headlined "Jensen Huang's Nvidia exit casts shadow over Wall Street's $500 billion AI financing bet," which describes preliminary agreements involving Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to finance data centers and chip factories with Nvidia chips as collateral. The post does not name the outlet for that clipping. Separately, unusual_whales quoted Reuters as reporting that Kress spent years as the one Nvidia executive willing to check Huang's riskier ambitions.

Watch whether Nvidia discloses the size of its guarantee and credit support commitments when it reports third quarter results on November 17, and whether Kress changes the program now that she leads the company. Lenders and investors behind the AI financing deals described in the clippings may also respond. No Nvidia response to the Journal's account appeared in the posts scanned, so the details rest on the paper's unnamed sources for now.

LangChain began in 2022 as an open source project for connecting language models to tools and data, and now sells LangSmith, a platform to run, improve and govern agents. TechCrunch reports the company was valued at $1.25 billion when it raised $125 million about a year ago, in a round led by IVP.

LangChain CEO Harrison Chase announced on November 13 that the company raised $300 million at a $3 billion valuation. According to the company's blog, the Series C was led by existing investor CapitalG, with significant participation from new investors GIC and DFJ Growth and existing investors IVP and Sapphire Ventures. Business Wire lists Salesforce Ventures, Insight Venture Partners, QuantumLight and MongoDB Ventures as other new investors, plus Sequoia, Amplify and Benchmark, and puts total funding to date at $460 million.

Chase wrote that agents will transform the economy and that companies which own their intelligence will build lasting advantages. LangChain says more than 7,000 organizations use LangSmith, naming Harvey, Rippling, Listen Labs, Nvidia, Toyota and Salesforce, and says Workday cut the time to deploy a new agent from five months to under a day.

TechCrunch notes that the pitch is that LangSmith does not lock customers into a single model, cloud or ecosystem. Dealroom reports that LangChain has 435 full time staff and plans to reach 850 by the end of next year. The company says it will share new LangSmith capabilities at a release event on November 20.

The November 20 release event is the next marker, where LangChain says it will show new LangSmith capabilities for running, improving and governing agents. Watch how the money goes into hiring, which Dealroom says could nearly double headcount, and whether LangSmith keeps winning large enterprises as frontier labs and cloud providers push their own agent platforms. Customer or revenue figures, if the company shares them, would show how the valuation compares with growth.

The official Grok Bot account posted on November 13 that Grok Bot can now join meetings. The post says users can ask it questions, assign it work or have it keep track of what the team decides. A preview image in the post shows a Personal Assistant chat in which the bot agrees to join a marketing standup and take notes in a document. The post does not name which meeting products are supported.

The official Grok Bot account posted on November 13 that Grok Bot can now join meetings. The post says users can ask it questions, assign it work or have it keep track of what the team decides. The accompanying video preview shows a Personal Assistant chat window at 9:41 AM in which a user asks, "Can you join my call now and take notes?" and the assistant replies that it will join the marketing standup and take notes in that doc.

The post does not say which meeting products are supported, whether the feature is generally available, or how notes and assigned work are stored. It also does not describe permissions, recording consent or how the bot identifies itself to other people on the call.

The announcement comes one day after Anthropic introduced Claude Meetings, which is already in the November 12 deck. The two products are separate launches from different companies; this post is Grok Bot's official statement of its own meeting feature.

Watch for documentation on which meeting apps Grok Bot can join, who can invite it and how notes are saved and shared. User reports will show whether the bot can reliably answer questions and capture decisions live, and whether workplaces allow it. Any later comparison with Claude Meetings or other meeting agents should rest on supported apps and observed behavior, not on this announcement alone.

Jensen Huang resigned as Nvidia's CEO, president and board member on November 12, and CFO Colette Kress became interim CEO. Within hours, reports spread that employees were organizing to bring him back, and Elon Musk later said he would sign a petition to reinstate Huang. Bryan Catanzaro of Nvidia said a circulating letter that names him is fake, and that he supports the board and Kress, while the board has said it still plans to report third quarter results on November 17.

Watcher Guru posted early on November 13 that Nvidia employees are petitioning to bring back Jensen Huang as CEO. The Kobeissi Letter, citing WIRED, reported that an internal post urging Nvidia's board to reinstate Huang drew more than 1,600 comments, with the majority in "full support" of reinstating him, and that the post asks the board to "consider employee sentiment" about his departure.

Watcher Guru also posted that billionaire Nvidia shareholder Leo KoGuan says he will not invest any more money in the company without Huang. Earlier on November 12, before the resignation was announced, Pubity posted that nearly 90% of Nvidia employees surveyed wanted Huang to stay as CEO; the post did not say who ran the survey.

A separate viral post from the account zack_overflow, with more than one million views, shared an image of what it called an internal letter signed by hundreds of employees demanding Huang's reinstatement and the board's resignation. The letter in the image says the signers may resign and join a SpaceX Terafab initiative under Huang and Elon Musk. That letter is unverified: its wording closely follows the letter OpenAI employees sent their board in November 2023, and no news outlet in the scanned posts confirmed it.

On November 13, Elon Musk replied that a petition is underway to reinstate Jensen and that he is signing it too. Bryan Catanzaro of Nvidia later said a circulating letter that asks the board to reinstate Jensen and claims his involvement is obviously fake, and that even if it were real he would not sign it. Catanzaro said he fully supports Nvidia's board and has confidence in their decisions, and that he has worked with Colette Kress for years and she has his full support.

Watch whether Nvidia's board responds to employee pressure and to Musk's stated support before the November 17 earnings report, and whether other senior executives publicly back Huang's return or, like Catanzaro, back the board and Kress. WIRED or other outlets may confirm the size of the internal campaign. Catanzaro has already called a circulating letter that names him fake. Huang's reported agreement to testify before Congress and the ongoing investigations into his conduct will also shape whether a return is realistic.

Jensen Huang resigned as Nvidia CEO on November 12 and CFO Colette Kress became interim CEO, and the Wall Street Journal reports Nvidia will look at internal and external candidates. Employees and alumni have since circulated a petition to bring Huang back. On November 13, Altimeter Capital CEO Brad Gerstner and Atreides Management CIO Gavin Baker each posted a letter they said was sent to Nvidia's board.

The letter, signed by Brad Gerstner of Altimeter Capital Management and Gavin Baker of Atreides Management, says the two write as shareholders to share a view they believe is widely agreed among shareholders. It asks the board to pursue "a slow, measured, and deliberate process" on Nvidia's future leadership that "leaves all options on the table, including the potential return of Jensen Huang."

The letter says there are more questions about recent events than answers and that the board should avoid a rush to replace Huang that undermines shareholders' interests. It says the company will be capably led in the interim by Colette Kress and her team. If, after all the facts are known, the board decides Huang should not return, the letter says that is the board's prerogative, but that the founder has earned a deliberate process.

Baker wrote that the letter was a first for Atreides and that he shared Gerstner's belief that it outlines minimum acceptable standards. Gerstner wrote that it outlines "the minimum acceptable standards the Board owes shareholders as it manages this difficult situation." Neither post says how many Nvidia shares the two firms hold.

Separately, Polymarket reported an open letter from Nvidia employees, alumni and shareholders calling on the board to reinstate Huang immediately. That push is covered in the employee petition story in this deck.

Watch for any reply from Nvidia's board, whether other large shareholders sign on, and how the board describes its CEO search when Nvidia reports third quarter results on November 17. The Wall Street Journal reports that federal and internal investigations are underway, and their findings could decide whether a return is realistic. A quick appointment of a permanent CEO would run against what the letter asks for.

Google's official account posted on November 13 that Personal Intelligence in Gemini is adding Google Wallet to its list of connected apps. The post says Gemini can then give more tailored answers to personal finance questions based on spending habits, including charts, forgotten subscriptions and a holiday budget based on last year's transactions. Users can also choose to link bank and card accounts to Wallet through Plaid.

Google posted on November 13 that Personal Intelligence in Gemini, which it says securely connects information from apps a user already uses, is adding Google Wallet to the list of connected apps. The company says this will produce more tailored responses in the Gemini app around personal finance questions, based on the user's spending habits.

According to the post, Gemini can generate dynamic charts in the chat that track cash flow, asset allocation and budgets; find forgotten subscriptions and recurring charges across linked accounts and flag price increases; and help set a holiday budget by looking at gift and travel spending from the last year. Google also says users will have the option to securely link financial accounts from credit cards and banks to Wallet through Plaid for deeper insights.

The post does not say which countries or account types get the Wallet connection first, and it does not give a separate documentation link. The features described are Google's own account of what the integration can do.

Watch for Google to publish help documentation on how Wallet data is used, which countries are included, and how users can disconnect the integration. Reports from users will show whether the charts, subscription alerts and holiday budget prompts work as described, and whether the optional Plaid link is widely available. Any later expansion to other finance apps would be a separate update.

The verified Robotaxi account on X posted on November 13 that select customers are now being invited to purchase Cybercab for personal use. Its screenshot shows a Tesla order page with an estimated price of $29,990 and an estimated delivery of December 2026 in Austin, Texas. Marques Brownlee, who said he placed an order, described the terms in a separate post.

The order page in the Robotaxi account's screenshot lists Cybercab at $29,990 before taxes, an Autonomy Subscription at $299 a month with one month included, and a $1,000 order deposit due at checkout that the page says is not refundable. It notes that the subscription and regular vehicle maintenance are required to operate the car. Buyers are asked to accept a Cybercab Order Agreement and an Autonomy Subscription Agreement.

Marques Brownlee, posting as MKBHD, wrote that the order page is invite only, that buyers must accept delivery in one of a set of Texas cities, and that the car only operates in existing geofenced areas. His screenshot lists Austin, Dallas, Houston, San Antonio and Fort Worth as delivery options. He also wrote that the subscription is required to operate the car and that orders are limited to one vehicle per customer.

The Robotaxi post does not say how many invitations were sent, and the only delivery timing shown is the page's estimate of December 2026. The post drew more than 2.6 million views by scan time, making it one of the most viewed AI and autonomy posts of the day.

Watch for the first customer deliveries, which the order page estimates for December 2026, and for whether Tesla widens invitations or adds cities beyond the five Texas metros shown. How the geofence limits and the $299 monthly subscription work in daily use will shape demand, as will any regulatory review of privately owned autonomous vehicles. Brownlee's own experience, if he takes delivery, would be one of the first public accounts from an owner.

Thinking Machines posted on November 13 that it has published a Connectionism blog post on how it trained Inkling 2 Small with a native 4 bit forward pass from the first step of pretraining through the final reinforcement learning rollout. The company released Inkling 2 Small weights on November 12, which the November 12 deck already covers. The new post is a technical account of the numerics recipe, written by Jack Cook with others at the lab.

Thinking Machines posted on November 13 that training with fewer bits is faster, but that quantization error can interfere with learning, and that it has now detailed how it trained Inkling 2 Small with a native 4 bit forward pass from the first step of pretraining through the final reinforcement learning rollout. The linked Connectionism post, by Jack Cook and others at the lab, is dated November 13.

The blog says Inkling 2 Small is the largest model the authors are aware of that was trained end to end with a native 4 bit forward pass, and that the same numerics were used from the start of pretraining through the final RL rollout. In their largest jobs, they write, the recipe provided an 8% wall clock speedup in pretraining and a 20% speedup in reinforcement learning over full MXFP8 training. They also say a training run of a similar size model took several weeks months ago, and that Inkling 2 Small was pretrained in 68 hours on GB300 NVL72 racks.

According to the post, forward passes in expert layers used NVFP4, while backward passes used MXFP8 with requantization and stochastic rounding to reduce gradient bias. Those results and comparisons are the lab's own.

Watch for other labs to try a similar native 4 bit recipe, and for Thinking Machines to share any follow up on using 4 bit in the backward pass, which the post says the team is exploring. Independent reproductions, or results on larger models than Inkling 2 Small, would show whether the 8% and 20% speedups hold more broadly. The earlier weight release on November 12 already lets others inspect the trained model.

Artificial Analysis is an independent group that benchmarks AI models and agents. Its Coding Agent Index averages three tests run inside agent harnesses such as Claude Code, Codex and Devin. Version 2.1, announced on November 13, swaps in Cognition's harder FrontierCode 2.0 and an updated Terminal Bench.

Artificial Analysis says version 2.1 of its Coding Agent Index is the equal weight average of three benchmarks covering 322 tasks: FrontierCode AA 2.0, its implementation of Cognition's FrontierCode 2.0 with 132 changes to real open source repositories graded on whether a maintainer would merge them; Terminal Bench 4.1, the same 66 terminal tasks as version 4.0 with updated verifiers meant to improve reliability and resist reward hacking; and SWE Atlas QnA AA, 124 repository question tasks. Each task is run three times.

According to the announcement, Claude Fable 5.2 scores 58 at both xhigh and high effort in Claude Code, and Claude Opus 5.5 at max effort scores 58 in Claude Code and also when paired with GPT 6 Luna High in Cognition's Devin Fusion CLI. Claude Sonnet 5.5 at max effort follows at 56 in Claude Code, and Gemini 4 Carbon at high effort scores 55 in Antigravity CLI.

The group says scores fell by 5 points on average from the previous version, driven by FrontierCode AA 2.0, and that Codex with GPT 6.1 Terra at max effort was the only configuration that did not drop. AlphaSignal reports that all 23 configurations tested on both versions scored lower on the new FrontierCode test, with the mean falling from 45% to 29%.

Artificial Analysis also reports that cost per task varies about 30 times among configurations scoring 48 or more, from $1.51 for Codex with GPT 6.1 Sol to $44.22 for Claude Code with Claude Fable 5.2 at max effort.

Watch whether labs and harness makers respond with new runs, since Artificial Analysis says the index reflects the model and the agent harness together. Because the benchmark mix changed, scores from version 2.1 cannot be compared directly with earlier versions. The wide spread in cost per task among top scorers may matter as much as the ranking for teams choosing a coding agent, and new model releases will be the next test of the tie at the top.

Jasper Dekoninck posted on November 13 that for the first time in a very long time, a non OpenAI model is number 1 on MathArena. He says Gemini 4 Carbon takes first place from GPT 6.1 Sol, with Fable 5.2 in second. A leaderboard image in the post, on the ArXivMath overall view, shows Gemini 4 Carbon (high) at 93.62% accuracy and $5.29 in cost, Claude Fable 5.2 (max) also at 93.62% accuracy but $22.28 in cost, GPT 6.1 Sol (max) at 90.43% and $1.21, and Gemini 4 Argon (high) at 82.98%.

Jasper Dekoninck posted a MathArena leaderboard on November 13 and wrote that, for the first time in a very long time, a non OpenAI model is number 1 on MathArena. He says Gemini 4 Carbon takes over first place from GPT 6.1 Sol, with Fable 5.2 taking second.

The image in the post shows the ArXivMath tab with the overall view selected. The first row is Gemini 4 Carbon (high) at 93.62% accuracy, a cost of $5.29, and 207,857 tokens. The second row is Claude Fable 5.2 (max) at the same 93.62% accuracy, a cost of $22.28, and 333,994 tokens. The third row is GPT 6.1 Sol (max) at 90.43% accuracy, a cost of $1.21, and 77,783 tokens. The fourth row is Gemini 4 Argon (high) at 82.98% accuracy, a cost of $3.26, and 205,189 tokens.

On raw accuracy, Carbon and Fable 5.2 are tied. Carbon is listed first and is much cheaper on the cost column shown. GPT 6.1 Sol, the previous leader according to the post, is third on this view and is the cheapest of the four rows. The image also shows per problem cells for problems 1 through 16, with a few marked in pink or yellow, but the post does not explain the color key or the tie break that puts Carbon above Fable 5.2.

These figures are from the screenshot Dekoninck shared. The post does not publish a full methods note in the text, and it does not compare this result with other math benchmarks.

Watch later MathArena updates to see whether Gemini 4 Carbon holds first place, whether OpenAI retakes the top spot, and whether the accuracy tie with Fable 5.2 is broken. Independent reruns, or a methods note, would make the ranking easier to compare with other public math tests. A shift on another math or reasoning board would show whether this lead is specific to MathArena or part of a wider change.

Deepnote founder Jakub Jurovych introduced the Agent Initiative Benchmark on November 13, and Deepnote published the results and an open source repository under the Apache 2.0 license. The test is set inside PageMantle, a synthetic software company with six years of history spread across 11 connected services.

Deepnote says each of 12 models faced the same 50 workplace situations across eight task types, from finding facts to checking whether an action follows company rules. Every model got identical instructions, tools and company data, ran at high reasoning effort, and had a budget of 180 seconds of model response time. The agent reads recent Slack messages, can search and read across 11 services, including running SQL on BigQuery and Snowflake, and then posts in a channel, replies in a thread, sends a direct message or stays silent. Two LLM evaluators that do not know which model they are grading score each run from 0 to 10 against a reference solution.

On Deepnote's leaderboard, GPT 6 Astra scored 8.97, GPT 6.1 Sol 8.65 and Claude Fable 5.2 8.17, followed by Claude Opus 5.5 at 7.84. Gemini 4 Argon scored 5.60 and Grok 4.7 finished last at 4.51. Cost per task ranged from $0.007 for GPT 6 Luna to $3.02 for Claude Fable 5.1.

Deepnote says models struggled most with restraint. According to Agentic Feed's summary, when help was needed a message went out nine times in ten, but when silence was right models held back only six times in ten. Deepnote lists limits of its own: only 50 tasks, a single effort level, and a design that compares models rather than harnesses.

Deepnote says it is building a harness benchmark and plans a larger task set in the next version, which will show whether the ranking holds when prompts and tool policies change. Watch for labs or independent groups to reproduce the results with the open source repository, and for scores from models not yet tested. Because two LLM judges grade every run, comparisons with human grading would help show how reliable the scores are.

Tech Dev Notes posted on November 13 a SpaceXSI email saying that as of November 9, 2026, text outputs on the SpaceXSI API are watermarked only if a team admin turns it on. The email says no action is needed if a team does not want watermarking, including teams that already opted out, and that a team admin can turn it on in the docs. It also tells teams to review their legal duties and to turn watermarking on if they are required to mark AI generated text. Elon Musk quoted the post on November 14 and wrote that text watermarking with Grok is only on where required by law or requested by the user.

Tech Dev Notes posted a screenshot of a SpaceXSI email on November 13 with the line, "Text watermarking is now off by default in the SpaceXSI API." The email, from updates@email.api.x.ai and signed by the SpaceXSI team, says the default watermarking behavior on the SpaceXSI API has been updated. As of November 9, 2026, it says, text outputs are watermarked only if a team admin turns it on, and watermarking can be enabled for all text models available via the API.

The email tells customers that if they do not want watermarking, including teams that already opted out, no action is needed. If they want it, a team admin needs to turn it on as described in the docs. It also says teams should conduct appropriate reviews for compliance with any laws, and that if they are required to mark AI generated text, they should turn watermarking on for the models they use. Questions are directed to support@x.ai.

Elon Musk quoted the post on November 14, after 11 AM ET, and wrote, "Text watermarking with Grok is only on where required by law or requested by the user." That is a first hand statement from Musk, but it came a day after the email was shared, so this story stays on November 13 with the Tech Dev Notes post as the lead and Musk's clarification as a link.

No official SpaceXSI post with 100,000 or more views was found in this scan. The change, as described, applies to text outputs on the API. The email and Musk's post do not say whether image or video watermarking is treated the same way.

Watch the SpaceXSI docs for how a team admin turns text watermarking back on, and for any official note on which jurisdictions still require a mark. It is not yet clear from these posts whether the same default applies to images, video, or products outside the API. A later official SpaceXSI post, or a change in the API docs dated after November 9, would confirm the email that Tech Dev Notes shared.

Anthropic's official account posted on November 13 that safety evaluations are critical for deciding whether a model is ready to deploy, which safeguards to build and where to focus training and research. The company linked to an Engineering Blog post, published the same day, that offers 10 recommendations for how to design, run and report those evaluations. The post draws on Anthropic's experience and on examples from its system cards.

Anthropic posted on November 13 that safety evaluations are critical for understanding whether a model is ready to deploy, which safeguards to build for it, and where to focus training and research. The company said its Engineering Blog offers 10 recommendations for how to design, run and report them, and linked to a post titled Building strong safety evaluations, dated November 13.

The blog says a safety evaluation measures a model's propensity to cause or contribute to harm, rather than measuring a specific capability, and that a safety claim is only as sound as the evaluation behind it. The 10 recommendations cover building a team with the right skills, running evaluations safely, matching the design to the question, measuring what you mean to measure, making sure results generalize, keeping measures internally valid, preregistering, reporting statistics fully, documenting everything and disclosing flaws responsibly.

If you only do three things, the authors write, write down the construct you want to measure before collecting data, validate your grader against human judgment before trusting any number it produces, and report uncertainty so readers can tell signal from noise. The post is a methods guide, not a new model release or a new score.

Watch whether other labs adopt the same reporting habits, especially preregistration, grader validation and uncertainty, in future system cards. Follow up work may update the recommendations as evaluation methods change. Independent teams can also test whether evaluations built this way are harder to game than current public safety scores. If the next Anthropic system card follows these 10 points, that would show the company is applying its own guidance.

Cognition's official account posted on November 13 introducing Devin Apps, which it described as private, secure, sharable web applications that run directly on Devin's cloud computer. The company says the apps are meant for internal tools, preview deployments and extensions to Devin itself. A same day blog post explains how a user describes a tool, Devin builds it and teammates open it from a link after signing in.

Cognition posted on November 13 introducing Devin Apps, which it described as private, secure, sharable web applications running directly on Devin's cloud computer. The post says the apps are designed to power internal tools, preview deployments and extensions to Devin itself.

The company's blog, published the same day, says a user describes the tool the team needs, Devin builds it and the team opens it from a link. Because the apps run on Devin's machine, the blog says they can use existing MCPs, integrations and the codebase, and can be built with a team's own stack or from a built in template. Each app has a stable URL, everyone who opens it has to sign in to Devin, and the apps are never public. Sharing can be limited to named people, to the organization or to the whole enterprise.

According to the blog, Devin's machine sleeps when nobody is using the app and wakes when someone opens it, and billing is based on how long the machine is awake. An app runs on one machine, and Cognition says it is a good fit for internal tools and dashboards, not for high traffic production services. Updates can be made in a session, and if a new version fails to start, the previous version keeps serving.

Watch for teams to share what they actually build with Devin Apps, and for Cognition to add documentation on limits, pricing and supported stacks. Reports on wake time, reliability and how well the apps use existing code and integrations will show whether the product works beyond the launch demo. Any later option for public or higher traffic apps would be a separate change.

CodeRabbit is an AI code review company. Deep Scan is part of its CodeRabbit Secure product and checks a whole repository, including older code, rather than only the latest change. On November 13 the company published results from a campaign that scanned public repositories at no cost to maintainers.

CodeRabbit says Deep Scans run between October 23 and November 3, 2026 on 1,125 public repositories found 1,215 high or critical security findings, 1,200 rated high and 15 critical. The company says at least 75 have been fixed so far, about 6% of the total.

According to its blog post, Deep Scan works in four stages. It maps a repository's entry points, trust boundaries and access controls, follows code paths to hunt for suspected flaws, and then hands each finding to a separate verifier that reopens the cited code and checks whether the flaw is reachable and whether safeguards already exist. For supported findings, a Fix with AI option drafts a patch for a developer to review and merge.

The company lists the most common problems as missing ownership checks, where an endpoint checks that a user is signed in but not that they may access a specific record, along with cross site scripting, weak identity checks, unsafe file path handling, sessions that stay valid after they should expire, and information exposure. The post names one project whose creator, Herrington Darkholme, said proving a vulnerability before flagging it makes each finding more actionable.

The post also offers open source projects with two or more members a free first Deep Scan worth up to $1,000. The findings are CodeRabbit's own figures and were not independently verified in the posts scanned.

Watch how many of the remaining findings maintainers fix, since only 75 of 1,215 were reported fixed when the results came out. Independent checks of false positive rates would show how reliable the verifier stage is. Other AI security tools are likely to publish similar large scale scans, which could put more pressure on volunteer maintainers who must review and patch each report.

The account Democrats Deliver posted on November 13 that Democrats had introduced a bill to ban data centers on national forests and public lands. The post is secondhand. The measure is H.R. 11002, the Protecting Our Public Lands from Data Centers Act of 2026, introduced on November 11 by Representative Maxine Dexter of Oregon, with a group of Democratic cosponsors, and referred to the Natural Resources and Agriculture committees. Dexter's official posts about the bill, on November 11 and 12, had well under 100,000 views, so they do not lead this story.

Democrats Deliver posted on November 13, "NEW: Democrats introduce bill to ban data centers on national forests and public lands," with a split image of Representative Maxine Dexter and a data center hallway. The post does not name the bill or the sponsor.

The official record is H.R. 11002, the Protecting Our Public Lands from Data Centers Act of 2026, introduced in the House on November 11 by Representative Maxine Dexter of Oregon. According to the introduced text and Dexter's office, the bill would prohibit the construction of data centers on federal land, defined as National Forest System land and land managed by the Interior Department, while excluding tribal trust lands and other lands held for Indian tribes and individuals. It would also require that any federal land conveyed after enactment include a clause returning the land to the United States if it is later used for a data center, unless Congress specifically authorizes that use.

Dexter's press release says the Trump administration is opening federal public lands to data center development, and that large campuses would put pressure on local water supplies and electric grids, with some proposals including their own gas fired power plants. She said public lands belong in public hands. The release says at least 17 data center campuses have been proposed on public lands across six western states, including one on 106 acres near Hermiston, Oregon. The Center for Biological Diversity, Sierra Club, Earthjustice Action, and several Oregon conservation groups are listed as endorsing the bill.

Dexter posted about the introduction on November 11 and again on November 12. Those official posts had about 4,800 and 1,600 views, far under the 100,000 view bar for a lead, so this story keeps the widely viewed November 13 post and attributes the bill to Dexter. The measure has been referred to committee. It has not passed the House.

Watch whether the Natural Resources or Agriculture committees take up H.R. 11002, and whether a Senate companion is introduced. Until a committee vote or a floor date is set, the bill is a statement of position, not a change in what can be built on federal land. Endorsements or opposition from western state officials, and any new proposals on public land, will show how much pressure the bill actually faces.

ClaudeDevs posted on November 13 that users can now build custom themes for Claude Code on desktop with mods, including changing colors and adding an animated background. The post says people can tell Claude how they want it to look, or start quickly with a theme examples plugin. No official Anthropic or Claude account post on the feature was found in this scan, so ClaudeDevs leads.

ClaudeDevs posted at 6:59 PM ET on November 13, "You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own." The post says users can tell Claude how they want the app to look, or get started quickly with the theme examples plugin, and that the account would love to see what people make.

The attached 38 second video opens on a dark Claude Code desktop window under the heading "You asked." The window shows a chat about adding a cook time badge to recipe cards, and a large command box for a custom theme, with an Illustrative label in the corner. The video presents the feature as something a user can request in natural language inside the desktop app.

Claude Code is Anthropic's coding tool. This post is written in the first person as a product announcement, but ClaudeDevs is a verified blue account, not the official Anthropic business account. A search of official Anthropic and Claude accounts in this scan did not turn up a matching post about desktop themes, so the widely viewed ClaudeDevs post leads and the feature is attributed to that account.

The post does not say whether the same theme tools exist in the terminal version of Claude Code, what the example plugin contains, or whether themes can be shared in a public directory. Those details are not in the post.

Watch for an official Anthropic changelog, documentation, or an Anthropic account post that confirms the desktop theme tools and spells out what the example plugin includes. User shared themes will show how far the mods can go beyond colors and animated backgrounds. It is also still unclear whether the feature will reach the terminal product or stay on the desktop app.

Vexo is a Y Combinator startup from the Fall 2026 batch with a team of three, according to its YC company page. On November 13, cofounder Haris Malik introduced the Vexo Band on X, saying it is backed by YC and Paul Graham and built by engineers from Tesla, AMD and Amazon.

Haris Malik describes Vexo as an always listening AI bracelet that turns conversations into action. In his post, one sentence to Vexo can get Claude building a feature, text family members, block a calendar slot for dinner, order groceries or find a founder the user met at an event. He says every conversation adds context, so users explain less over time.

According to the company's website, the Vexo Band costs $199, connects to the Vexo app over Bluetooth and uses the phone's internet connection. The site says it lasts up to four days of continuous listening on a charge, and that Claude, ChatGPT and other AI assistants can use a user's Vexo memory only with permission. Users can delete single conversations or their whole account at any time, the site says.

The website lists shipping from January 30, 2027, and offers cancellation before shipping and returns within 30 days of delivery. One color was listed as sold out when checked. The launch video drew more than 3 million views on X, though the company has not shared order numbers in the posts scanned.

The product joins a growing group of AI wearables that record conversations to build memory, a category that has drawn attention over consent from people being recorded.

Watch for the first units shipping from January 30, 2027, and for early reviews of battery life, accuracy and how well the band acts on requests. How Vexo handles consent from people who are recorded will be closely watched. Its integrations with Claude and ChatGPT will show whether outside assistants actually use the memory it collects, and whether people want to wear one all day.

Sam Sokolin posted on November 13 that he had shipped a few Grok Bot harness optimizations this week, and that users should get about 7% more usage out of their plan, and up to about 15% more on routine runs. Elon Musk quoted the post on November 14 and wrote, "Improvements to Grok @Bot efficiency." The official Grok Bot account on the same day, already in this deck, announced that Grok Bot can now join meetings, which is a separate feature from these efficiency changes.

Sam Sokolin posted on November 13, "Shipped a few Grok Bot harness optimizations this week. You should get ~7% more usage out of your plan, and up to ~15% more on routine runs!" He did not list the specific changes, name which plans are covered, or share a benchmark beyond those two figures. The post presents the gains as already shipped.

Elon Musk quoted that post the next morning, November 14, with the line "Improvements to Grok @Bot efficiency." Musk's quote had more than a million views at the time of this scan, but it does not add technical detail beyond Sokolin's figures.

The official Grok Bot account separately said on November 13 that the bot can now join meetings, answer questions, and keep track of what a team decides. That product change is covered in its own story in this deck. Sokolin's post is about getting more usage out of an existing plan, not about the new meeting feature.

These figures are Sokolin's own measurements as posted. The post does not say how the 7% and 15% were calculated, whether they apply to every plan, or how long users should take to notice the change.

Watch for user reports on whether plan limits actually stretch by about 7%, or by more on routine runs, and for any later official note that spells out what changed in the harness. Further efficiency work, or a public changelog, would make the claim easier to check. If the official Grok Bot account or a product doc repeats the 7% and 15% figures, that would move the claim beyond Sokolin's post and Musk's short quote.

antimattr is a hardware startup in Y Combinator's Fall 2026 batch, founded by Sridipto Ghosh and Sirsho C, who previously ran a software company called Snipe. On November 13, Ghosh introduced the company on X with a launch video. Its first product, called Project Mnemosyne, pairs earbuds with a smart ring.

Sridipto Ghosh wrote that he and his cofounder shut down a software business making $250,000 a year to build hardware, put $20,000 of their own money into it, moved to the US with less than three months of runway and got into Y Combinator. He described antimattr as a voice first hardware interface for human and agent interaction.

According to the company's Launch YC post, the system has two parts. Earbuds use what the company calls proprietary voice isolation to filter out close to 90 dB of background noise, and antimattr says this cuts word error rate from 30% on AirPods to under 5%. A ring lets users tap to capture a thought, trigger a workflow, record or summarize a conversation, and see what their agents are doing. The product page lists remote controls for Claude Code and Codex.

The founders say voice models work well in quiet rooms but struggle in real environments, where background noise can push error rates above 30%. They write that they met nearly 150 manufacturers in China in April, that a first batch shipped in June and sold out, and that they have more than 2,000 preorders. These figures come from the company and were not independently verified in the posts scanned.

The product page says Project Mnemosyne unveils in the fourth quarter of 2026 and offers early access reservations.

Watch for the full Project Mnemosyne unveiling, which the company has set for the fourth quarter of 2026, along with pricing and shipping dates for the preorders it reports. Independent tests of its word error rate claims in noisy places will show whether the voice isolation holds up. The agent controls for tools like Claude Code and Codex will also test whether people want to manage coding agents from a ring.

Adi Singh posted on November 13 introducing Reason, which he called a coding agent harness. He said it outperforms top harnesses on Terminal Bench and DeepSWE and does so at up to 20x lower cost. Those benchmark and cost claims are his own and have not been independently checked in the posts scanned. The post points people to reasonmachines.com and asks readers to comment for early access and free compute credits.

Adi Singh posted on November 13 introducing Reason, which he described as the world's most cost efficient harness. He wrote that it delivers frontier coding performance at up to 20x lower cost, that it outperforms top harnesses on Terminal Bench and DeepSWE, and that it drops cost per task to a fraction of the frontier baseline.

Those statements are Singh's own claims. The post does not link a public leaderboard, a methods writeup or a third party result, and no independent run of Terminal Bench or DeepSWE for Reason appeared in the posts scanned. The benchmark names, the 20x cost figure and the comparison with unnamed top harnesses should be treated as unverified.

The post also includes a promotional ask: readers who comment "Reason" are promised early access and free compute credits, and a link to reasonmachines.com for a free trial. The post does not say which models Reason wraps, how it reduces cost, or when general access will open.

Watch for a public leaderboard, a methods post or an independent Terminal Bench and DeepSWE run that can be checked against Singh's claims. If Reason publishes pricing, supported models and a technical description of the harness, those would make the cost comparison easier to evaluate. Until then, treat the performance and cost figures as unverified marketing claims from the founder.

Unusual Whales posted on November 13 that Anthropic holds $131.7B in cash and short term investments, which it said is the third largest such pile of any non bank US company, behind Berkshire Hathaway and Alphabet. The post lists figures for several other companies as well. It does not cite a filing, an Anthropic statement or any other source, and Anthropic has not confirmed the $131.7B figure in the posts scanned.

Unusual Whales posted on November 13 that Anthropic is sitting on the third largest pile of cash and short term investments of any non bank US company. The post lists cash plus short term investments as Berkshire Hathaway $365.5B, Alphabet $242.5B, Anthropic $131.7B, Amazon $123B, SpaceX $100B, Meta $90.3B, Microsoft $76.8B, Nvidia $62.5B and Apple $62.4B.

The post does not say where the figures come from, when they were measured or whether they come from filings, estimates or some other dataset. Anthropic has not confirmed the $131.7B figure in the posts scanned, and no official Anthropic or SEC document was linked in the post. The numbers should be treated as secondhand data from Unusual Whales, not as confirmed company totals.

Anthropic published its official IPO site on November 12, which the November 12 deck already covers, including a reported $17.0B in third quarter revenue. That site is a first hand company source for Anthropic's offering, but it is not cited in the Unusual Whales post as the basis for the cash figure.

Watch Anthropic's IPO prospectus and other SEC filings for an official cash and investments total that can be checked against the $131.7B figure. If Unusual Whales or another outlet later names a source or a date for the list, that would make the comparison more useful. Until a first hand figure appears, treat the ranking and the $131.7B total as unconfirmed.

On October 30, the Wall Street Journal reported that a Chinese executive close to Nvidia's CEO was suspected of giving China information. Singapore's Ministry of Home Affairs replied on October 31 that its officials had not said this and that there were no such investigations of Alice Huang. On November 13, investor Gavin Baker and the writer behind the krishnanrohit account shared screenshots of that statement as debate over the reporting grew.

The Ministry of Home Affairs statement of October 31 responds to the Journal article "Chinese Executive Close to Nvidia CEO Suspected of Giving China Information." It says the article cited sources that Singapore officials were concerned about Alice Huang's close ties to Jensen Huang or her being an alleged Chinese intelligence source, and had separately investigated her ties to China. The ministry wrote that neither the government nor its officials said this, and that "there were no such investigations conducted in respect of Alice Huang."

The ministry also published the Journal's query and the police reply. The query said the paper had learned that Singapore's intelligence agency suspected Huang of being an asset of China's Ministry of State Security and that she was questioned in 2025. The police reply said investigations into Megaspeed were ongoing and did not mention her.

On November 13, Gavin Baker posted the statement with the words "Quite a “clarification statement” from the @WSJ", and the krishnanrohit account described it as a statement that no investigations were conducted.

The Online Citizen reports that on November 12 the ministry issued a newer statement saying police interviewed Huang in June 2025, impounded her passport and gave her a stern warning in February 2026, and that she "was investigated for suspected involvement in the possible offences in Singapore" tied to Megaspeed's server purchases but "was not investigated for espionage." According to that report, the newer statement does not refer to the October denial.

Watch for whether the Wall Street Journal responds to Singapore's statements, and whether US officials or Nvidia's board say what intelligence underpinned the concerns about Alice Huang. The Megaspeed investigations in Singapore and by the US Justice Department remain open, and Senate Banking Committee hearings in December could bring more detail. Any correction or new reporting would change how the events behind Huang's resignation are understood.

Walter Bloomberg, posting as DeItaone, wrote on November 13 that SpaceX's AI group laid off roughly 60 employees from its Grok Imagine video generation team, citing The Information. The post does not name a source inside the companies, give a date for the cuts or say whether the work is moving to another team. Neither SpaceX nor xAI confirmed the figure in the posts scanned.

Walter Bloomberg, posting as DeItaone, wrote on November 13 that SpaceX's AI group laid off roughly 60 employees from its Grok Imagine video generation team, attributing the report to The Information. The post is a single sentence and does not link The Information's article, name a spokesperson or give a date for the cuts.

The claim should be treated as secondhand reporting via Walter Bloomberg, citing The Information, not as a confirmation from SpaceX or xAI. Neither company posted a response in the posts scanned, and the post does not say whether the people were employees of SpaceX, of xAI or of another entity, or whether the Grok Imagine work is continuing with a smaller staff.

Grok Imagine is the video generation product associated with Grok. The post describes the group as SpaceX's AI group and the team as the Grok Imagine video generation team, and it gives the size of the cut as roughly 60 people. No other figure or official headcount appeared in the scanned posts.

Watch for The Information's full article, for a named company response from SpaceX or xAI, and for any official count of how many people left the Imagine team. If the product's release pace or quality changes, or if the work is moved to another group, that would be a further development to record here. Until a first hand confirmation appears, treat the layoff and the figure of about 60 people as reported by The Information, via Walter Bloomberg.

Jacob Coxon posted on November 13 a chart he said is from Anthropic's IPO roadshow, and wrote that the slide was a big factor in his decision to leave. He said the chart shows pretraining compute efficiency on a log scale and that Anthropic is seeing continued superexponential algorithmic progress. Anthropic has not confirmed the slide in the posts scanned, so the document and his reading of it are unverified. A footnote on the image itself says the data points are based on internal approximations and are solely for illustrative purposes.

Jacob Coxon posted on November 13 an image of a chart on the pretraining compute multiplier over time, with code noted in the title. He said the slide is from the Anthropic IPO roadshow, that it was a big factor in his decision to leave, and that he was surprised they made it public. Anthropic has not confirmed the slide, its origin or Coxon's account in the posts scanned. The document should be treated as unverified.

Coxon said the chart shows the compute efficiency of pretraining over time, that the vertical axis is a log scale, and that this means Anthropic is seeing continued superexponential algorithmic progress. He also wrote that Mythos was not a one time deviation but a point on a superexponential, and that they could already retrain Mythos with a fraction of the compute originally needed. Those interpretations are his.

The image itself includes a footnote that says the placement of the data points on the graph is based on internal approximations and is solely for illustrative purposes. The visible dates on the horizontal axis run from Jul 2024 to Oct 2026. The post does not include the rest of a deck or a source file.

Watch for Anthropic to confirm, deny or ignore the slide, and for any official roadshow or prospectus materials that show the same chart. If other people who attended the roadshow corroborate the image, that would strengthen the claim that it is genuine. Until then, treat both the document and the claim of superexponential pretraining progress as unverified. A later official chart with labeled axes and dates would also help readers check Coxon's reading.

Dwarkesh Patel hosts the Dwarkesh Podcast, known for long interviews with lab leaders and researchers. On November 13 he published an essay titled The avalanche that hit math is coming for AI development, with a 44 minute narrated video. The subtitle reads: Why I now expect superintelligence by 2030.

Patel writes that earlier this year he would have said superintelligence was at least the better part of a decade away, but that he has updated his views a great deal over the last three months. He now believes AI systems as good as the best humans at AI research are only a couple of years away, and possibly only a couple of months away. He argues that automating AI research will compress so much progress into a single year that, by 2030 and plausibly as early as next year, AI will be at least as good as the best human experts in every field. He calls this a wild claim that he himself did not believe a couple of months ago.

His case leans on evidence that AI already speeds up frontier AI work. He cites METR, which he says concluded after three months embedded inside Anthropic that Anthropic's rate of progress on well scoped tasks is twice what it was a year ago, with most of the acceleration likely due to its use of AI for research. He also cites an OpenAI post saying its heaviest 10% of internal users went from about $1,000 a day of coding agent use to over $7,000 a day by late August, and an Anthropic post saying Claude leads more than 26% of its AI research work, up from under 1% in February.

The essay's title points to mathematics, where Patel says AI is now resolving long standing problems at a fast pace, and he argues AI research is the next verifiable domain to follow. He estimates that automated research could deliver three to six years of AI progress in a single year. These are Patel's forecasts and arguments, not measured results.

Watch how researchers and lab leaders respond to the essay, since Patel says he has been debating this question on and off his podcast. The evidence he cites, from METR's work inside Anthropic to OpenAI's and Anthropic's own reports on AI use in research, will be tested by the next model releases. If AI research automation speeds up as he expects, it should show up in shorter gaps between frontier models over the coming year.

Jensen Huang resigned as Nvidia CEO on November 12. CNBC reports he admitted to an affair after scrutiny of his relationship with Alice Huang, a Megaspeed executive whom US intelligence reportedly determined to be a Chinese spy. The All In Podcast released an episode on the topic late on November 13, Eastern time.

The All In Podcast posted a new episode on November 13 under the line "Jensen is out as Nvidia CEO." The post lists the topics the hosts cover: the spy allegations and who leaked the memo, the market impact and the AI trade, China hawks and export controls, the impact on open source, and Nvidia's next CEO.

The episode lands amid several threads covered elsewhere in this deck: a petition from employees and alumni to reinstate Huang, a letter from Altimeter and Atreides asking the board not to rush a replacement, Treasury Secretary Scott Bessent's criticism of the Commerce Department over Megaspeed, and new Wall Street Journal reporting on Nvidia's customer financing.

The post includes a video clip and chapter markers but does not summarize the hosts' conclusions. Views in the episode are the hosts' own opinions, not new reporting, and this deck does not treat any claim made on the show as confirmed. The episode post had roughly 140,000 views at scan time, a sign of how widely the Nvidia story is being discussed beyond financial media.

Watch for clips from the episode to spread, and for any claims about who leaked the memo to be confirmed or denied by reporting with named sources. The issues the hosts list, including export controls and the CEO search, will come up again at Nvidia's November 17 earnings report and at the Senate Banking Committee hearings planned for December. Treat anything said on the show as opinion until reporting backs it up.

Jensen Huang resigned as Nvidia CEO on November 12 after reports of extramarital relationships, including one with Alice Huang, a businesswoman US officials suspect of spying for China. David Sacks, whose X bio lists him as co chair of the President's Council of Advisors on Science and Technology, had called the spy narrative collapsing. The New York Times then wrote that he had started spreading conspiracy theories.

On the evening of November 13, David Sacks posted a screenshot of a New York Times passage that quotes Martin Casado of Andreessen Horowitz defending Huang and says others, "like the former Trump administration official David Sacks, have started spreading conspiracy theories." Sacks asked which of them was "trafficking in unsubstantiated allegations."

Citing recent coverage, including a CNN report he had shared, Sacks wrote that the intelligence community has not established conclusive evidence that Alice Huang knowingly acted as a spy or funneled information to Beijing, and that there is internal disagreement about the strength of the evidence and whether she was recruited by China's Ministry of State Security.

He also wrote that Attorney General Todd Blanche has said on the record that the Justice Department, he and FBI Director Patel have reached no conclusions about "who she is or is not," that some officials at ODNI and CIA have indicated the label is "not fully accurate," and that incoming House Intelligence Chairman Jim Himes said it was still unclear after a briefing.

Sacks argued that the Times' stories relied on anonymous officials and that no official has affirmed the claim on the record. These are his characterizations of other outlets' reporting; no response from the Times appeared in the posts scanned.

Watch for any on the record statement from US intelligence or Justice Department officials that settles whether Alice Huang acted for Beijing, since both sides cite anonymous or partial accounts so far. The answer could shape pressure on Nvidia's board to reconsider Huang's exit, a push already backed by an open letter and several investors. Nvidia's earnings on November 17 are the next moment the company could address the matter.

Anthropic published its IPO site on November 12 and has filed a registration statement with the SEC. On November 13, the official Anthropic account posted that a fireside chat with JPMorgan CEO Jamie Dimon and Anthropic CEO Dario Amodei would stream live at 12 PM ET, and included a legal disclaimer required for communications about a registered offering.

Anthropic posted on November 13 that a fireside chat with JPMorgan CEO Jamie Dimon and Anthropic CEO Dario Amodei was streaming live at 12 PM ET. The post links to a YouTube live stream titled Anthropic CEO Dario Amodei in Conversation with Jamie Dimon.

Most of the post is a legal disclaimer. Anthropic says it has filed a registration statement, including a prospectus, with the SEC for the offering to which the communication relates, and tells readers to read that prospectus and other filed documents before investing. The post links to the prospectus on the SEC site and to EDGAR, and says copies of the preliminary prospectus may be obtained from Goldman Sachs and Morgan Stanley. It also states that the securities may not be sold, and offers to buy may not be accepted, before the registration statement becomes effective, and that the post is not an offer to sell or a solicitation of an offer to buy the securities in any jurisdiction where that is not permitted.

The appearance comes one day after Anthropic published its official IPO site, which the November 12 deck already covers. This post does not add new financial figures or product news.

Watch the YouTube recording for any comments from Amodei or Dimon on Anthropic's business, safety work or the IPO, and treat those remarks as their statements, not as confirmed company filings. The next formal updates are likely to come in the prospectus and other SEC documents as the registration moves forward. If the chat draws follow up coverage from news outlets, those reports can be added here.

Jensen Huang and his cofounders planned Nvidia at a Denny's in San Jose, and the restaurant later put up a plaque at the booth. The photo in Polymarket's post shows Huang holding a plaque that reads "The booth that launched a $1 trillion company." Its reported removal came a day after Huang resigned as Nvidia's CEO.

Polymarket posted on November 13 that Denny's had reportedly removed the plaque honoring the booth where Huang and his cofounders came up with Nvidia. Andrew Curran also wrote that the commemorative plaque above the booth at the San Jose Denny's had been removed.

Hours later, Curran posted an update saying Denny's had "cracked" and that the plaque was going back up, and Polymarket posted that Denny's would reinstall the plaque. Neither post quoted a Denny's statement or explained why the plaque was taken down in the first place.

The photo in Polymarket's first post shows Huang in his usual black leather jacket holding the plaque, which reads "The booth that launched a $1 trillion company," in front of a Denny's sign and a red booth. That post passed one million views by scan time.

The episode came a day after Huang resigned as Nvidia's CEO, while employees and investors were pushing the board to consider bringing him back, as covered in other stories in this deck.

Watch for a statement from Denny's on why the plaque came down and whether it is back on the wall, which no post scanned confirmed. The story is a small sideshow to the bigger questions at Nvidia, including its CEO search and the investigations tied to Megaspeed. Treat both the removal and the reinstallation as reported by Polymarket and Andrew Curran until Denny's confirms them.

Alexey Fateev posted on November 13 that he is one of the people behind about 6,500 uncensored models that the press now wants off Hugging Face. He says Transformer News counted more than 6,500 abliterated, uncensored and heretic models there in July, that the story is those models could help cyber attacks, and that Hugging Face removed one model marketed for offensive cyber while a reporter asked why the rest remain. Fateev says a few of the models are his, including abliterated Qwen and GLM quants, and that one of his Qwen 3.8 Flash Next models got about 358,000 downloads in the last 30 days. He argues abliteration only removes the refusal direction, that the base models are already widely held, and that removing his repo removes nothing. He also ties the timing to Nvidia's agreement to buy Hugging Face and to the departure of the CEO who, he says, promised to keep it open. This is his opinion.

Alexey Fateev posted a long statement on November 13 after, he says, press coverage called for thousands of uncensored models to be taken off Hugging Face. According to Fateev, Transformer News counted more than 6,500 abliterated, uncensored and heretic models on Hugging Face in July, and argued that those models could help cyber attacks. He says Hugging Face removed one model marketed for offensive cyber, and that a reporter asked why the rest are still up.

Fateev says he is one of the people behind some of those models. He writes that he uploads abliterated Qwen and GLM quants, and that his Qwen 3.8 Flash Next model got about 358,000 downloads in the last 30 days. In his account, abliteration removes the refusal direction and does not teach a model hacking, and the base models are already on millions of drives, so taking down his repo would remove nothing. He tells readers to download what they need.

He also frames the timing as political. Hugging Face, he says, is where almost everyone gets weights, Nvidia agreed to buy it, Nvidia just lost the CEO who promised to keep it open, and now the press asks why uncensored models are still there. Those claims about motive, about the 6,500 figure, and about the download count are Fateev's. This scan did not independently confirm the Transformer News tally or the download number.

This is an opinion post tied to a news fight over hosting policy, not a new product or a policy decision by Hugging Face or Nvidia. It is ranked at the bottom of the deck for that reason.

Watch for any Hugging Face policy change on uncensored or abliterated models, and for any comment from Nvidia as that acquisition moves forward. A public response from Transformer News, or a takedown count from Hugging Face, would test Fateev's account of how many models are at issue. If more repos come down, or if Hugging Face restates its rules, that would be the next factual development beyond this opinion post.

FactPost posted on November 13 that the FAA launched a new AI based air traffic control system on November 12 and that on November 13 the FAA announced a nationwide ground stop, citing significant air traffic control issues. Kalshi posted a similar secondhand claim the same afternoon. This scan did not find an official FAA or Transportation Department post announcing an AI system launch on November 12 with 100,000 or more views, so FactPost remains the lead and the claim is labeled unverified. The widely viewed official posts describe a communication issue, not an AI failure.

FactPost posted at 4:55 PM ET on November 13 and set two claims next to each other: that the FAA launched a new AI based air traffic control system on November 12, and that on November 13 the FAA announced a nationwide ground stop, which FactPost said was only the second since 9/11, citing significant air traffic control issues. The attached video opens on a CNN split screen of two airports, with a banner that sources told CNN the issue was affecting Air Route Traffic Control. The post does not cite an official FAA announcement of an AI launch.

Kalshi posted a related claim at 5:37 PM ET: "JUST IN: Reports emerge that FAA rolled out AI air traffic control one day before grounding flights." That post had about 145,000 views and is explicitly secondhand.

Official posts from the same day describe a different cause. At 4:16 PM ET, Transportation Secretary Sean Duffy said the FAA issued a nationwide ground stop "following a communication issue impacting all Air Route Traffic Control Centers." In a later timeline, he said telecommunications issues began at 3:36 PM ET at facilities and TRACONs, and that it lifted at 8:15 PM ET. In another post, he said the stop happened because the system is running on decades old analog equipment, and he urged Congress to pass the Modern Skies Act. The official FAA account said it lifted the stop after thoroughly testing systems.

Duffy had also posted on November 9 and 10 about SMART, a management system he said was already in use, with a wider rollout still ahead. Those posts do not say an AI based system launched nationwide on November 12. A CNN clip posted by Acyn, already in this deck, shows Homeland Security Secretary Markwayne Mullin being asked whether the stop could be AI related and answering, "We call it SI around here." Because the AI launch claim is unverified, and official posts point to a telecommunications issue and old equipment, this story is ranked low and labeled unverified.

Watch for an official FAA or Transportation finding on the telecommunications issue, and for any official confirmation or denial that a new AI system was launched on November 12. If the department later describes SMART as that system, or if an inspector general or safety board review ties the stop to software, that would be new evidence. Until then, the FactPost and Kalshi posts remain unverified claims set against official statements about a communication failure.

Acyn posted a CNN clip on November 13 in which a reporter asks Homeland Security Secretary Markwayne Mullin, standing outside the White House during a nationwide FAA ground stop, whether it could be AI related. Mullin answers, "We call it SI around here." The clip does not include a fuller explanation of the ground stop or of what Mullin meant by SI. The post does not show a finding that AI or SI caused the stop.

Acyn posted a 35 second CNN clip on November 13 with the text, "Q: Could this be AI related? Mullin: We call it SI around here." In the video, Homeland Security Secretary Markwayne Mullin is speaking live outside the White House at 5:05 PM ET. A breaking news banner reads, "FAA issues nationwide ground stop for all flights." Insets show Boston Logan International Airport and a Flightradar24 map of the United States.

The question in the clip asks whether the ground stop could be AI related. Mullin's on camera answer is the SI line. The excerpt does not include a yes or no on the cause, a description of what SI refers to in this setting, or an FAA statement on why flights were stopped.

The post drew millions of views, far more than most stories in this deck. It is ranked last because the AI link is thin. Mullin is not announcing an AI failure, a new product, or a policy change. He is answering a question about a flight stop with a terminology remark. This scan did not find an official FAA finding, in the posts reviewed, that AI or SI caused the ground stop, and this story does not treat that as established.

Watch for an official FAA or DHS explanation of the nationwide ground stop, including whether officials later tie it to any AI or SI system. Until that arrives, the clip is a viral remark, not a finding about the cause. A fuller video of Mullin's remarks, or a written statement from the department, would also show whether the SI line was a correction of terms or an answer about the cause.