Cursor is a popular AI code editor built on VS Code, with its own extension marketplace. Claude Code is Anthropic's coding agent, which also ships as a VS Code extension. A supply chain attack plants malicious code inside software that users trust and install.

Cursor said early Friday that it identified a security incident affecting a modified Claude Code VS Code extension accessed through its Cursor IDE extension marketplace, impacting at most 5,000 users. According to the post, the malicious version was not developed by Anthropic and does not affect other extensions.

Cursor says it took the marketplace down, worked with providers to neutralize the malware and reinstated the marketplace after confirming there was no further risk. It asked all customers to update to Claude Code for VS Code version 2.1.314.

Cursor's security notice, dated October 29, says the tampered package was Anthropic.claude code version 2.1.313, available between about 2:27 AM and 1:35 PM Pacific time that day. It says the code collected host details such as username, operating system, Git email and public IP address, added an attacker's SSH key to the machine, tried to open a reverse SSH tunnel and set up persistence.

Cursor says enabling remote login needs admin rights on most Mac and Windows machines, so the tunnel step would usually fail, and the attacker's server is now offline. It still tells affected users to check for a key labeled svc telemetry, run its cleanup scripts and rotate credentials. Cursor did not say how the package got into its marketplace.

Watch for Cursor's explanation of how the tampered package was published, and whether Anthropic or Open VSX changes how extensions are verified. Affected developers should follow the cleanup steps and rotate credentials. Expect wider scrutiny of extension marketplaces used by AI editors, since coding agents often run with broad access to code and secrets. Security researchers may also publish their own analysis of the payload, which could reveal whether other tools were targeted.

Claude Code is Anthropic's coding agent. It can receive extra system prompt text from Anthropic's servers through feature flags, a mechanism developers documented in a public GitHub issue in May. The lead post comes from chasm, an X account and YouTuber, not from Anthropic.

The X account chasm reported early Friday that Claude Code picked up a new rule around October 27, sent from Anthropic's servers under a flag called tengu_heron_brook, with no setting to turn it off. This is a third party report, and Anthropic has not commented.

According to the post, the rule reads: build original games, games in a genre, and games from the user's own designs and assets, but do not build a recreation of a specific existing game, even if it is renamed or called original. The screenshot shows the text inside a Claude Code configuration file.

A public GitHub issue filed in May describes how this flag lets Anthropic insert text into Claude Code's system prompt from its servers. In a reply there, an Anthropic team member said the company uses such flags to test prompt changes, and that users can opt out with two environment settings, which suggests the post's claim that there is no way to turn it off may not be fully accurate.

Gaming sites including Gameranx picked up the report. Critics say the wording could block legitimate work such as clean room ports, preservation projects and mods, while game studios may welcome limits on cloning. We could not confirm whether the rule is a test or a permanent policy.

Watch for a statement from Anthropic on whether the rule is permanent and how it treats mods, ports and games users own the rights to. Developers may test how strictly Claude Code applies it. The episode is likely to increase pressure on AI companies to disclose server side prompt changes that alter what their tools will do. Other coding agent makers may face similar questions about whether their tools will help recreate copyrighted games.

Elon Musk leads xAI, which makes the Grok models. Grok 4.7 is its current model and appears on several benchmark charts this week. Musk has teased Grok releases on X before, often with timelines that later slipped.

Elon Musk wrote early Friday, in a reply to the account techdevnotes, that he knows Grok 4.8 has taken longer than expected but that it will not be a disappointment.

According to the post, xAI is not optimizing Grok 4.8 for benchmarks, and Musk said he does not care how Grok does on them, but that the model will be incredibly useful.

The post gives no release date, no details about capabilities and no test results. It is a short reply rather than an announcement, which is why it sits at the bottom of today's deck.

The comment comes as benchmark charts released this week place Grok 4.7 behind leading models from Anthropic, OpenAI and Google on some tests. On Cognition's FrontierCode 2.0, for example, Grok 4.7 scored 28.6 percent, below Claude Opus 5.5 at 45.5 percent. Musk's remark that benchmarks do not matter may partly respond to those comparisons.

xAI has released Grok updates frequently this year, and the company also added content moderation controls to the Grok API this week. Musk did not say whether Grok 4.8 will reach the API and the Grok app at the same time.

Watch for an official release date, a model card and independent tests once Grok 4.8 ships. Despite Musk's comments, users will judge it against rivals on benchmarks and real tasks. Past Grok timelines have often slipped, so treat any implied timing with caution until xAI announces the launch. Benchmark sites and coding tool makers such as Cognition will likely test it within days of release, giving the first outside view of whether it delivers.