
Democrats and Republicans in Congress are backing measures and bills to further restrict exports of AI chips following a scandal involving former Nvidia CEO Jensen Huang and a suspected Chinese spy. Lawmakers have demanded that Huang and other Nvidia leaders testify, and Senate Minority Leader Chuck Schumer says Huang and the Trump administration must answer to Congress. Senate Banking Committee Chairman Tim Scott has sent letters to Huang and Commerce Secretary Howard Lutnick asking them to testify publicly.
TechCrunch reports that China's most capable AI models were trained or run on Nvidia chips that are subject to U.S. export restrictions. The piece suggests the restrictions have not fully cut off Chinese AI labs from top-tier Nvidia hardware. The source excerpt contains only the headline and link, so details on how the chips were obtained and which models are affected are not confirmed here.
OpenAI's blog describes a candidate nonperturbative lattice definition of the Standard Model of particle physics. The company says its model, GPT-6 Astra, helped develop the idea into a detailed proof. The post's description gives few technical specifics, so the claim's significance depends on independent review of the proof.
Anthropic released Claude Fable 5.2 on November 9 and claims it beats Claude Opus 5.5 on almost every capability. The New Stack's comparison finds the two models post the same scores, but Fable 5.2 costs almost four times as much to run.
MIT researchers report that OpenAI's GPT-6 Astra edited about 0.016% of a small Qwen model's weights, with no gradient descent or training data, removing a habit of repeating until it ran out of context. The model's MATH score rose from 7.9% to 48.1%, but most of the gain came from fixing the repetition: math answers hitting the length limit fell from 93% to 33.1%, not from new math knowledge. The three arXiv papers from November 6 are still awaiting independent reproduction, and Matthew Honnibal has argued the claims are overstated.
Uber Engineering argues that traditional identity and API controls are insufficient for AI agents that reason over external content and take real-world actions. To address this, Uber built a framework that evaluates whether an agent's action is appropriate given the user's request, the agent's intended purpose, its permissions, and the interaction context.
Bharat Semi, an Indian chip startup, has closed a $24 million seed round. The company says it has won design wins for its AI-enabled sensor technology with customers including General Atomics. It is now targeting the global AI sensor market.
Anthropic's initial public offering has launched, according to CNBC, and it carries significant weight across the tech industry, which holds large stakes in the AI lab. The listing arrives as AI stocks have been under pressure and investors are looking for positive momentum. The source excerpt does not include pricing, valuation, or trading details.
Nvidia's security chief left the company quietly in August, according to TechCrunch, and CEO Jensen Huang has since stepped down over personal relationships, including one with an executive at a company suspected of funneling chips to China. A former intelligence officer quoted in the report says the episode reflects a kind of information leak that most corporate security teams are poorly equipped to catch. The source text is brief, so the details of the departures and the leak itself should be checked against further reporting.
The U.S. Treasury Secretary questioned whether Nvidia knowingly sold AI GPUs to shell companies and said the company itself bears responsibility. He stated, "I think it is important that anyone at the company there be held accountable for that."
Anthropic has redesigned Claude Projects so a coordinator agent scopes a task, delegates it to parallel Claude Code cloud sessions, and assembles the results. Each thread runs on its own copy of the repo and Git branch, can spawn subagents, and keeps working after the user closes their laptop. The public beta is rolling out gradually to select Pro, Max, and Ultra subscribers, with Team and Enterprise plans excluded for now and a waitlist for others.