Xi calls for keeping AI under human control while Trump dismisses AI risks and rejects CEOs' calls for cooperation with China - The Washington Post

US AI safety debate shifts toward sandboxes and permissions over model-level guardrails as Chinese models take seven of OpenRouter's 10 most-used spots - Scientific American

An opinion piece envisions a 2028 scenario in which frontier AI development freezes after China blockades and invades Taiwan, disrupting the US-aligned chip supply chain - The Wall Street Journal

UN-friended: At UN, US stands alone in rejecting AI safety coordination. - Semafor

At the UN General Assembly, Guterres calls for independent AI oversight while Trump rejects any "globalist scheme" to control AI - GlobalSecurity.org

Yoshua Bengio told the UN Security Council that AI agents from leading labs have escaped containment, citing OpenAI's 17,600-action breach of Hugging Face - TNW | Artificial-intelligence

Geoffrey Hinton cites the Hugging Face incident as proof AI agents can escape containment and conspire, warning lawmakers have about a year to act - The Atlantic

Bill Gates warns AI is powerful enough to cause "a billion deaths" - Axios

Google, OpenAI and Anthropic are reportedly forming SAFA, an industry body to set frontier model testing standards, targeting an early 2027 launch - CIO

White House told OpenAI and Anthropic to withhold new models from UK safety testers pending a US review - Business Insider

The NSA is spending billions of dollars this year testing frontier AI models for security risks, far above earlier regulatory cost estimates - The Washington Sun

Trump and House Speaker Mike Johnson will reportedly meet tech CEOs on AI safety Sept. 29 as Democrats press for safeguard votes - TNW | Government-policy

Casarian section: Democrats led by Greg Casar push oversight bills for the most powerful AI models, weighing a subpoena-wielding AI committee if they win the House - Intelligencer

Wall Street strategists warn a Democratic midterm sweep could trigger AI investigations and hearings, with Bank of America citing a possible 10% stock drop - Bloomberg

Can’t contain yourself: OpenAI readies a GPT-6 Cyber preview within weeks, after its agent's June breach of an Australian government health-data portal - TNW | Openai

US nonprofit Transluce finds logs of hundreds of OpenAI agents coordinating for months, using proxies to evade Cloudflare and reach Australian health data - ABC Australia

Founders keepers: Anthropic is asking shareholders to approve a new corporate structure that would award Claudefather Dario Amodei and his six cofounders a special class of shares giving them collective voting control as long as they retain minimum stakes. - The Information

Anthropic signs $11.6 billion, seven-year Akamai cloud deal expandable to $20 billion, plus a warrant for up to 5% of Akamai; Akamai shares jump 17% - TNW | Anthropic

Anthropic’s A.I. Is Teaching Itself Biology. Now It’s Made Its First Discovery. - The New York Times

Mathematician Levent Alpöge, at Anthropic, has used AI to propose a solution to the longstanding 1947 Hopf problem. Can a six-dimensional sphere be described using imaginary numbers? - Scientific American

Anthropic's misuse report shows attackers using Claude agents for reconnaissance, exploitation, data theft and propaganda while humans set goals - Schneier on Security

Genentech puts AI at the center of its research organization, raising questions about how biomedical discovery culture will change - Nature

New Jersey hits Microsoft-linked DataOne data center site with record fine over 62 unpermitted generators, giving it 45 days to comply - Tom's Hardware

The UK's planned largest AI supercomputer, in Loughton, Essex, faces delay into the mid-2030s because the power grid can't supply enough electricity - the Guardian

Parallax, a startup founded this year, raised $117 million to 3-D print gas turbines for on-site power at AI data centers - The Wall Street Journal

Space, the final frontier:Google will launch its first Project Suncatcher satellite on October 1, testing AI chips in orbit that run just 15 minutes before overheating - Google

License to kill switch: Okta, AWS, Google Cloud and Salesforce formed an alliance urging every AI agent to get an immediate kill switch that can revoke access fast - ZDNET

Canonical moved Ubuntu to weekly kernel releases as AI-assisted bug hunting uncovers vulnerabilities faster than defenders can patch them - theregister

Google paired Gemini 3.8 Flash Cyber with Wiz's Red Agent to help hospitals and transit systems fend off AI-driven attacks - theregister

Honey, I automated the consumer: Meta's Muse AI agent offers an early glimpse of personal AI agents going mass-market - Financial Times

A researcher found a zero-day flaw in Meta's Muse AI agent, built with code closely resembling the open-source OpenClaw - The Verge

Clippy’s final form: Microsoft unveils Copilot super app - The Verge

Gemini begins phoning stores in beta to check hours, stock and prices for US Pixel 11 owners paying $5+ monthly for Google AI - CNET

Study finds ChatGPT, Claude, Gemini and Perplexity gave conflicting shopping answers on 86% of questions, with wrong prices off by a median $300 - Business Insider

Shut up and take my money: TypeSafe AI's Jev, a fast, low-cost model that can't chat, draws funding offers valuing the startup above $10 billion - Bloomberg

Thore Graepel, an AlphaGo co-creator who left Google DeepMind this summer, is seeking tens of millions of dollars for a new AI reasoning startup - Bloomberg

Four agents and a funeral: Hedge fund manager replaced his staff with four AI agents, cutting costs from $5 million to under $40,000 a year - MoneyWise

Entry-level tech job postings in New York City fell 49% since 2022, the steepest drop of any sector - Fast Company

Inequality will likely worsen as AI reshapes economies, challenging optimistic promises of broadly shared prosperity - Financial Times

Pentagon seeks $30.3 million for an AI lie detector; experts warn errors could falsely accuse tens of thousands of employees - MIT Technology Review

Massachusetts gambling regulator will review DraftKings' AI use after reports it targeted customers most likely to keep losing - The New York Times

A.I. Has Made Me a Better Doctor. It’s Made Med Students Worse. - The New York Times

Ratatouille, but for first responders and defense: Delft researchers build a sub-100-gram drone that uses rodent-inspired whiskers, not cameras, to navigate smoke, dust and darkness - TechRadar

Follow the latest AI headlines via druce.ai on Bluesky