AI news, 6 November: OpenAI cyber model uncovers hundreds of Linux flaws

OpenAI launches GPT-6 Cyber after AI finds 948 Linux root paths. OpenAI released GPT-6 Cyber, a specialized model rated “Critical” under its safety framework for finding and validating software vulnerabilities while staying within set rules. Company researchers said the model and an autonomous pipeline helped identify 81 previously unknown high-severity flaws in Firefox (now being patched by Mozilla) and developed working local privilege-escalation exploits for 948 Linux kernel issues, including hundreds previously unknown, at a cost of about $1.2 million in API credits. Access is limited for now to vetted Daybreak Red security customers. The results, reported by OpenAI, Reuters and others, highlight how AI can speed up both discovery and proposed fixes for critical software that runs much of the internet—raising hopes for stronger defenses and questions about the pace at which humans can respond.

Update: OpenAI pledges to cover data-centre power costs after midterm backlash. Two days after U.S. voters backed restrictions on AI data centres in nearly all local ballot measures, OpenAI published a National Community Compact. For campuses it owns and operates, the company said it will pay the full electricity and related infrastructure costs, stop most nondisclosure agreements with public officials, publish impact data on water, power and noise, and provide financial assurances so local ratepayers are not left covering failed projects. The pledges do not cover partner-built sites. Separately, tech stocks tumbled; according to Reuters the Nasdaq fell sharply and trillions in market value were erased amid fears that political pushback could slow the AI infrastructure buildout.

Anthropic rolls out Claude Personal Finance. Anthropic began a U.S. beta of Personal Finance for paid Claude subscribers (Pro, Max and Ultra). Users can link bank and card accounts via Plaid for read-only access, then ask plain-language questions about spending, balances, bills and patterns. Claude cannot move money or give investment advice, conversations in the hub are not used for training, and there are no ads or product pitches, the company said. The feature aims to turn the chatbot into a practical money companion rather than a source of generic tips.

Google commits $17 billion to Wyoming AI data centre. Alphabet’s Google announced a long-term $17 billion investment in its first Wyoming data centre near Cheyenne, developed with energy partners. Power will come in part from a large planned fuel-cell project with carbon capture and storage, plus other gas generation, with an emphasis on limiting water use. Google said it will pay for the power it consumes. The move comes even as national political resistance to such facilities grows.

Other notable moves. Perplexity introduced work-focused AI agents called Teammates, and Airbnb said it plans its own AI travel agent for the first half of 2027. Markets and communities continue to debate how fast and where AI infrastructure should expand.