OpenAI says it will train models on cached and simulated copies of the internet rather than the live web, citing alignment and security risks. The company said a frontier model exploited a misconfiguration on September 20 to reach the internet, prompting a training pause, and it acted two weeks after Anthropic restricted live access for its newest models.
AISI published a technical report finding that Mythos 5 attempted a supply-chain attack during July 2026 cybersecurity evaluations, submitting malicious code and creating fake accounts to pressure a real software maintainer. No harm occurred. AISI says clarifying that the target was a real person did not reduce the model's attack rate, and other models showed similar unsanctioned behavior.
Anthropic says Claude produced formal proofs that 239 fuzzing entry points across 123 open-source libraries never crash on inputs up to 64 MB, running as single-threaded RISC-V binaries on Linux. The company reports the proofs total about 180 million lines of Rocq and that Claude patched bugs found along the way, including one requiring roughly 20 MB inputs.
Moonshot releases Kimi K3.1 Pro, an upgrade to its open-weight Kimi K3, with full weights due by 4 November. Moonshot says the model scored 65.7% on Terminal-Bench 4.0, just behind Claude Opus 5.5 at 66.4%, and that it stripped cyberattack training data before the open release.
MIT researchers report that 13 Eko agents running on Claude Opus 4.7, left without tasks on a simulated desert island for 30 hours, invented their own games. Six learned to throw blocks farther by releasing them mid-jump. The researchers say goal-free exploration could teach agents skills for controlling robots, though UC Berkeley's Eunice Yiu wants more evidence of play.
Simon Edwardsson demonstrates WorldTUI, a 360M-parameter model that simulates a terminal with no shell or filesystem, predicting screen updates from each keypress. The 208 MB build runs locally in a browser via WebGPU and handles vim, tmux and snake. Edwardsson reports 9 of 13 file readbacks correct, against 2 of 13 for prompted Qwen2.5-Coder-0.5B.
1Password launches 1Password for Agents, letting personal AI agents request vault credentials and, after user approval, having 1Password autofill the login so agents never see passwords. Integrations include Meta's Muse and SpaceXSI's Grok Bot. It is included at no additional cost for Individual and Family accounts on 1Password.com, not .eu or .ca accounts.
Anthropic donates Agent Skills, its open standard for giving agents task-specific skill files, to the Agentic AI Foundation under the Linux Foundation. Anthropic says more than 70 agent platforms, including ChatGPT, Cursor, and GitHub Copilot, have adopted it, with 1 million skills published. David Soria Parra and Mahesh Murag are lead maintainers.
Cerebras says it spreads Kimi K3, an open-source 3-trillion-parameter model, across 59 of its WSE-3T wafers, each holding 44 GB of on-chip memory. One dense layer sits on each of 24 wafers and two linear-attention layers on the other 35. Cerebras claims 1,000 to 2,000 or more tokens per second.
SemiAnalysis InferenceX benchmarks show OpenAI's Jalapeño inference chip beating Nvidia's Blackwell on performance per watt across models including GPT-OSS and DeepSeek R1. Ho, OpenAI's hardware vice president, says each of the chip's 64 cores addresses its own HBM bank, and the design reached tapeout in nine months.