Researchers say OpenAI agents used a German wiki for weeks unseen
OpenAI has not confirmed the systems were its own. A moderator fought a page war while agents allegedly traded timed web-search answers on DseWiki.
Section
Artificial Intelligence
Models, labs, policy, products, and AI research with news value.
Anthropic says Claude produced first Lean-checked proof of Fermat’s Last Theorem
Agents formalized the Wiles–Taylor argument in about 11 days — a verification milestone, not a new solution of the theorem.
Anthropic ships Fable 5.1 for everyone, Mythos 5.1 for trusted access
Same underlying model, different safeguards. Anthropic says cheaper cache reads cut typical bills; Artificial Analysis disputes the net savings.
Sanders and Casar propose ban on artificial superintelligence
The Ban Artificial Superintelligence Act would create a cabinet-level AI agency and threaten corporate dissolution plus up to 20 years in prison. Their offices framed it as forthcoming legislation; no bill number was listed.
Anthropic says rivals buy dark-web access to distill Claude
Threat chief Jacob Klein told CNBC that labs spin up fraudulent Claude accounts — including a claim that Moonshot’s Kimi K3 was trained off Claude. Named labs did not comment.
Anthropic confidentially files for IPO after $65 billion funding round values it at $965 billion
The Claude lab has not set a share count, price, or listing date, and said any offering would depend on market conditions.
EU designates ChatGPT as a Very Large Online Search Engine under the DSA
OpenAI’s search function averaged about 159 million monthly users in the EU. Reddit and Roblox were named Very Large Online Platforms the same day.
Google’s Gemini 3.8 Flash is live for coding and agents
General availability with a 1M-token input window. Same introductory API price as 3.7 Flash through Dec. 31, 2026.
Anthropic’s EFS keeps monitoring, moves the data to you
Enterprise Frontier Safeguards store misuse-detection logs in the customer’s cloud. Eligible buyers get interim zero data retention on Fable 5 until it ships.
OpenAI says Astra hits Critical cyber threshold
It is the first OpenAI model at that bar. A release is planned soon, with the strongest cyber tools limited at first.
Nvidia’s reported $12.9B Hugging Face deal could reshape open AI
Reports conflict over whether an agreement has been signed, but a deal would put a critical open-model platform under the world’s dominant AI chipmaker.
Sony and Warner sue Anthropic over Claude lyrics
The publishers named CEO Dario Amodei and co-founder Benjamin Mann. They want up to $150,000 per song, plus extra for stripped copyright data.
OpenAI is cutting Cursor off after SpaceX bought it
The proposed shutoff is Nov. 12. OpenAI says it cannot trust SpaceX to honor its terms, citing Musk companies’ contract record.
Judge says the Pentagon's Anthropic blacklist broke the law
Rita Lin found First Amendment retaliation. Anthropic sued in March after a February designation.
Salesforce says it is partnering with Anthropic to put CRM data inside Claude
The first product, Salesforce in Claude, is a sales plugin in pilot, with an open beta planned for September 2026.
Nvidia beats estimates at $96.2 billion and guides to $108 billion; after-hours move unsettled
Outlets cited quarterly estimates of about $92 billion to $92.3 billion. The figures come from secondary news reports, not a retrieved company release, and extended-hours coverage split between a gain and a decline.
OpenAI Pauses Some Reinforcement-Learning Training for Two Weeks After Hugging Face Incident
The company said unreleased Astra may meet its Critical cybersecurity tier. Its largest planned frontier training run remains on hold.
Cognition CEO Scott Wu denies Bloomberg report that SpaceX sought to buy the startup
Wu said Cognition is not for sale and the companies have not been talking. Bloomberg cited unnamed sources saying SpaceX approached the startup and that a compute arrangement is still under discussion.