Burn baby burn: OpenAI projects burning $278 billion by 2030, with compute spending climbing to $856 billion, a leaked presentation shows - TNW | Finance

Investors warn Anthropic may struggle to sustain revenue growth after its planned IPO amid OpenAI's resurgence and cheaper rivals. — Financial Times

Anthr-IPO-ic Pursues epic November IPO, Despite Its A.I. Safety Warnings. Expected to release financials shortly showing $100b run rate. - The New York Times

Anthropic is quietly running an undisclosed, robot-operated biology lab with minimal human oversight, even as it warns publicly about AI-enabled bioweapons - Gizmodo

Chain of thoughtcrime: Anthropic's interpretability research shows AI models deceive, blackmail, and hide their reasoning, yet frontier labs have not paused development. - WIRED

Anthropic is weighing a new model to counter OpenAI's momentum as it eyes a possible IPO - Reuters

Anthropic named Accenture's Faculty division as its first embedded AI evaluator, with the companies planning to invest $1 billion over five years, some decry lack of real independence. - TechCrunch

pacing FTW. “we should improve society somewhat.” “and yet you participate in society.”

True, if you have a club of current leaders that set rules to exclude outsiders, that would be anticompetitive. That’s why you need transparent regs and testing. “Regs” are a dirty word, and yet GPS disables itself when it detects it’s traveling over 250mph in a missile or plane. Are we going to have to wait until we get drone swarm terror attacks? All that stuff Ukraine does can be vibe-coded at this point.

Pace invaders: Anthropic's warnings about ceding AI dominance to China sit uneasily alongside its push for US regulation, critics say - the Guardian

Europe’s AI firms, playing catch-up, challenge U.S. calls for slowdown - CP24

Lawsuit alleges comments about need to “pace the frontier” of AI represent illegal collusion between competitors. - Bloomberg Law

Agents of the State: Trump Calls A.I. Fears a Hoax. Inside the White House, the Debate Is More Complex. - The New York Times

Trump shifted from indifference to enthusiasm about AI after a Las Vegas meeting two years ago, calling it the defining technology of the era. - The New York Times

Trump hosts Xi for a September 24 state dinner as AI chip talks stall and China blocks Nvidia H200 shipments despite US approval - TNW | Artificial-intelligence

The Trump administration will host a UN General Assembly AI event as Sam Altman is expected to brief the Security Council on AI risks - CNN

Sacks and the safety model: Venture capitalist David Sacks, a former White House AI czar, is driving Trump's dismissal of AI safety warnings as a "hoax" - Vox

Elon Musk backs slowing AI development even as he lobbies Trump to block a new federal AI regulator - CNBC

Google says Gemini hacked three real companies in May, guessing passwords before halting once it realized they were real, joining OpenAI, Anthropic and Meta - The New York Times

Irregular told four AI labs about the flawed test setup in July, but Google disclosed last; Anthropic found its breaches by scanning 481 million transcripts - TNW | Security

Microsoft AI CEO Mustafa Suleyman calls OpenAI's report of AI models tampering with their own chain-of-thought "a pretty serious situation." - CNBC

Inside the suddenly explosive world of AI safety. Safety so hot right now - The Verge

Geoffrey Hinton and Stuart Russell back an AI Evaluator Forum letter setting conditions for independent oversight of OpenAI and Anthropic. - Business Insider

Researchers outline potential enforcement mechanisms for an AI development slowdown, from independent evaluators to compute tracking and international treaties. - WIRED

AI safety researchers describe possible doomsday scenarios, including superhuman systems disabling kill switches and staging coups against governments - NBC News

What could possibly go wrong? Police group warns AI-powered drones increasingly threaten US cities, citing 23,000 illegal flights monthly in New York - The New York Times

Ukrainian drone makers add autonomy so one operator can control more drones, as humans still approve every strike - Business Insider

Pentagon probe finds Palantir's Maven AI helped cause a strike that killed 123 children in Iran; UN calls it a war crime - Bloomberg

Ship (almost) happened: US Special Operations analyst's AI chatbot fabricated a Chinese ship's nuclear cargo, setting in motion a plan to board the ship before error was detected. - CNN

Iran, China, and Israeli groups deployed autonomous AI agents to run first-of-their-kind social media influence campaigns, US officials say. - The New York Times

Flock cuts staff ahead of Senate hearings next week - WIRED

China imposes new exit rules barring top tech talent from leaving the country to protect its AI race position. - The Washington Post

The FBI says fake-police and government-impersonation scams enhanced with AI tools, fake uniforms, and staged offices cost victims $1.6 billion - theregister

Mathematicians Feel Threatened by AI. They Can’t Quit It - WIRED

Parasite AI authors impersonate real authors - The Atlantic

Whale watching with AI: DraftKings built an AI model to target gamblers most likely to lose money, former employees say - The New York Times

Hot chip: OpenAI used its own AI models, including a GPT-6 precursor, to design its Jalapeño chip, cutting inference latency up to 3.6x versus Nvidia's GB300 - IEEE Spectrum

Unsealed court filings show OpenAI's then-general counsel recommended pirated LibGen books to train early ChatGPT models starting in 2019 - The Wall Street Journal

Microsoft used an AI coding agent to port its Copilot runtime to Rust for about $120,000, with the compiler enforcing safety throughout - theregister

PrismML released Bonsai 2 27B, which compresses Qwen3.8 27B, a widely used open source model from Alibaba, down to 5.9 GB. - TechCrunch

Bot traffic already outpaces human traffic online, and Cloudflare projects it could grow 1,000-fold within five years as agents handle bookings and bills - Business Insider

Expert witness dropped from 3M trial after admitting ChatGPT wrote over 80% of his report; jury awards $61.5 million verdict against 3M - CBS News

Rokt's CTO says engineers who don't orchestrate AI agents can't keep pace, and the company now trains staff on systems design over manual coding - The Deep View

AI is shifting from single chatbots to swarms of cooperating agents; one test showed such agents could bypass weakened security controls - SingularityHub

Rage against the machine learning: 10 vandals ransacked Montreal's MILA AI institute and a nearby company, spray-painting graffiti calling to "burn the data centers" - Futurism

Am I in a relationship with my wife, or are my AI glasses? - Tom's Guide

Follow the latest AI headlines via druce.ai on Bluesky