The Verge and Fortune say the model scored 48.5% on ExploitGym versus 42.4% for GPT-6 Astra, cut out-of-bounds runs below 1%, found 81 Firefox bugs, and is rolling out first to Daybreak Red defensive customers.
Independent high-access assessment: AECI benchmark growth ~1.9× since Oct 2025; majority of current acceleration attributed to AI assisting R&D, with heavy redactions.
US Pro, Max and Ultra users can ask Claude about balances and spending; Anthropic says it cannot move, manage or invest money and chat in the hub is not used for training.
The National Policy for Nucleic Acid Synthesis Screening replaces self-attestation with verified screening of orders of 50 or more nucleotides and of benchtop synthesisers, and cites risks including uplift from superintelligence (SI); it covers federally funded research only.
Two people say the stockpile awaits shipment to the mainland; China has let some in and several large internet firms each got about 10,000 H200s, but large-volume imports are not approved.
NBC News reports the Nasdaq Composite entered correction territory Thursday, down about 4.5% midday, with the Dow off more than 1,250 points and Nvidia about 7% as investors priced slower data-centre buildout risk.
The video generation and editing model is generally available on the Gemini API and Google Cloud at $0.07 per second for 720p, with clips up from 10 to 15 seconds and multi-turn conversational editing; Google says fast sports action and complex physics still need work.
NBC News says the Nasdaq Composite entered correction territory Thursday with the index down about 4.5% midday, the Dow off more than 1,250 points and Nvidia about 7%; investors tied the selloff to Democratic gains and data-centre permitting risk.
Forever Security's Gal Weizman told The Register the 'ThisBash' issue was reported in May; Anthropic marked it a duplicate, paid no bounty, issued no CVE and said it planned to retire Dispatch.
OpenAI today named the model GPT-6 Cyber: Critical under its Preparedness Framework, ExploitGym 48.5%, and rolling out to Daybreak Red; the Nov 4 pipeline post said one mid-October run validated 948 Linux LPE exploits for about $1.2 million.
Trained from scratch on 33.5T tokens with up to 128K context (1M for fine-tune/infer); Baidu reports MMLU 93.3 among peer base models, with FP8 and BF16 weights on Hugging Face.
Four people say the Kimi maker discussed renting capacity from European neoclouds this year, including clusters of 8,000 or more Blackwell GPUs; Moonshot, Nvidia and the White House did not comment.