Regeringskansliet, the Government Offices of Sweden that supports the Prime Minister and cabinet ministries, manages sensitive policy records subject to strict Swedish public secrecy and data sovereignty laws. Earlier in 2026, Regeringskansliet restricted DeepSeek and awarded a procurement contract to Swedish startup Intric to run a sovereign internal AI assistant platform on domestic infrastructure.

Sweden's Government Offices, known as Regeringskansliet, will block browser access to OpenAI's ChatGPT and Anthropic's Claude on staff work computers starting Monday, October 26, according to primary reporting by Swedish public broadcaster Sveriges Radio Ekot and public television network SVT Nyheter. Polymarket amplified the news on X on Sunday morning, though its one line summary omits several nuances confirmed by the Swedish government's press office.

In an internal intranet notice obtained by Sveriges Radio Ekot, Regeringskansliet digitalization director Mikael Wettercrantz stated that using external AI services without a formal contract carries consequences for the government's information security. Instead of open consumer chatbots, civil servants are directed to use Microsoft Copilot, which is available to all employees and recently had its security settings tightened following an internal review. However, in a written statement to SVT Nyheter, Regeringskansliet clarified that while the public web interfaces for ChatGPT and Claude are being blocked on staff computers, selected employees retain access to both ChatGPT and Claude language models through Intric, a sovereign internal AI platform hosted on Swedish infrastructure.

Sveriges Radio's investigation shows the cabinet restriction is part of a broader clampdown across Swedish state agencies. The Swedish Tax Agency, Skatteverket, and the National Board of Institutional Care, Statens institutionsstyrelse, have already enacted similar network blocks on unapproved AI tools, while the Swedish Enforcement Authority, Kronofogden, is preparing its own decision. Earlier in 2026, an internal audit at the Swedish Transport Administration, Trafikverket, warned of widespread and risky staff use of roughly ten open AI services involving sensitive agency data.

Other Nordic and European Union ministries will watch Sweden's dual track approach of pairing hardened Microsoft Copilot access with a domestically hosted multi model platform like Intric. Meanwhile, the Swedish Enforcement Authority is expected to announce its own AI access decision soon, and enterprise AI vendors seeking public sector contracts across Europe will face growing pressure to sign formal government data processing agreements or deploy models inside sovereign cloud environments.

Grok Bot is xAI's autonomous personal agent platform, which provides users with a persistent Linux workspace, browser automation, and app connectors. Over the past week xAI has rapidly expanded Grok Bot's capabilities, rolling out a credential Vault and iterating through version 0.77 prerelease builds.

Elon Musk announced on Sunday morning that Grok Bot can now place outbound phone calls to individuals or businesses on a user's behalf, make a specific request during the call, take notes on the other party's response, and send the user a written summary of the conversation. Musk quoted an earlier post by developer Alex highlighting new capabilities spotted in Grok Bot version 0.77.

According to the 0.77 build details summarized by Alex and prerelease tracker blankspeaker, the new Calls panel allows a user's bot to dial a number, sit on hold with a dedicated status indicator, press keypad tones to navigate automated phone trees with a pressed keys readout, and handle call transfers while the user watches a live real time transcript of the conversation. This allows the agent to handle routine telephone tasks such as booking appointments, checking business hours or inventory, and waiting through customer service queues.

Beyond outbound voice calling, the 0.77 build updates highlighted in the quoted thread include several system and connectivity additions. Grok Bot can now join a user's private Tailscale network once approved by a tailnet administrator, and users can designate a default tailnet for their bot. The update also introduces a Payments settings screen that layers on top of the recently launched Vault feature to connect payment accounts and choose a default payment method, an early setup screen for Apple Messages on Mac under Connect Apps that remains switched off for now, and email invitations for sharing a bot with other Premium or Ultra subscribers.

Users and developers will test how reliably Grok Bot handles complex interactive voice response menus, regional accents, hold music, and live receptionists in real world calls, as well as what rate limits and safeguards xAI applies to prevent automated spam calling. Observers will also watch when the Apple Messages on Mac integration and connected payment workflows move from prerelease flags into general availability.

For years Nvidia has segmented its consumer GeForce graphics cards from higher priced workstation and data center GPUs by halving the instruction issue rate of FP16 and BF16 Tensor Core operations that accumulate into FP32 precision. While physical silicon on Turing, Ampere, and Ada Lovelace GPUs contains full rate tensor units, firmware and register locks cap their throughput on GeForce models.

Open source hardware modders have published Linux kernel modules that remove Nvidia's artificial throughput caps on FP16 and BF16 Tensor Core operations with FP32 accumulation across consumer GeForce RTX 20, 30, and 40 series GPUs, as reported by VideoCardz and summarized on X by Pirat_Nation. Rather than soldering hardware resistors or flashing a custom VBIOS, the tools run at Linux boot in the initramfs before the stock Nvidia driver loads, using an analysis of Nvidia's debug booter to override streaming multiprocessor speed select configuration registers. The register override survives PCIe Function Level Reset and standard CUDA initialization, and resets cleanly whenever the GPU is powered off.

The effort builds on developer Zagitta's GPU Unlock project for Ampere RTX 30 cards, TOTORONG's AD103 unlocker payload, and JustAResearcher's GPU Unlock Ada repository for Ada Lovelace GPUs. According to community benchmark figures reported by the modders, register level instruction throughput for FP16 with FP32 accumulation on an RTX 4090 doubles from 165.6 to 331.6 TFLOPS, with BF16 rising from 165.5 to 330.6 TFLOPS. On Ampere, modders report an RTX 3090 Ti rising from 87.5 to 157.4 TFLOPS in BF16 and an RTX 3090 rising from 73.0 to 122.8 TFLOPS in PyTorch GEMM, while developer KohakuBlueleaf demonstrated a Turing RTX 2080 Ti climbing from 59.9 to 111 TFLOPS.

Independent testing highlighted on DEV Community notes an important distinction between raw instruction issue rates and end to end workloads. While microbenchmarks show double the tensor instruction throughput on an RTX 4090, real cuBLAS GEMM matrix multiplications gain 12 to 34 percent in FP16 and 16 to 20 percent in BF16 depending on matrix shape. On an RTX 3090, LLM prompt processing improves by roughly 14 to 36 percent, whereas token generation remains largely unchanged because it is bound by memory bandwidth, and gaming frame rates are unaffected.

Local AI enthusiasts running 24GB RTX 3090 and RTX 4090 workstations will test the Linux kernel modules on real LLM prompt prefill, LoRA fine tuning, and image diffusion workloads to verify long term thermal and numerical stability under stock power limits. The hardware community will also watch whether Nvidia blocks the pre driver register override in upcoming Linux driver releases or whether modders extend the technique to RTX 50 series Blackwell cards.