Down from Cloud 9: Growth slowed to about 45% quarter over quarter, and rising compute costs pushed Anthropic back to a $567 million operating loss - Business Insider

Anthropic's amended IPO filing shows nearly $17 billion in Q3 revenue, up twelvefold, and $895 billion in compute and infrastructure commitments - The New York Times

Buy side-eye: Institutional investors tell Anthropic its IPO valuation should be $1.5-1.6 trillion, below earlier $2 trillion estimates, after a slowdown in hypergrowth and a tech stock selloff - The Wall Street Journal

Critics question the IPO's lofty valuation and the $582 billion non-cancelable share of compute commitments, with one urging the SEC to block the listing - theregister

Gigawatt expectations: Inside Anthropic's quest for compute - The New York Times

Anthropic's new Claude Enterprise "smart reports" tie team AI spending to work output and flag no-value or off-topic sessions - Business Insider

Field of DRAMs? Sam Altman calls scarce DRAM sitting in rarely used PCs "quite terrible"; Elon Musk mocks him as memory shortages raise device prices - Tom's Hardware

frankly my dear, I don’t give a DRAM

OpenAI partners with Cathedral Systems for cybersecurity and improves bug bounties. - OpenAI

One quota to rule them all: OpenAI will merge ChatGPT and Codex usage limits into a single allowance for Plus, Pro and Business subscribers, drawing user backlash - TechRadar

Charles Porch Is OpenAI’s Hollywood Whisperer - Intelligencer

Researchers use OpenAI's GPT-6 Astra to edit Qwen's weights directly, fixing a reasoning loop by changing 0.016% of parameters without retraining - DeepFabric

Get out the volt: Data centers shape Ohio and Michigan races, with candidates in 20 of 36 governor's contests backing moratoriums - Fast Company

Sorry, Mother Earth, $70b data center project would account for 4% of Canada's emissions. - Futurism

Realtor.com study of 43 ZIP codes finds new data centers didn't significantly dent nearby home values, though it excluded Texas - Tom's Guide

Apple's new Siri AI app saves chats; users can bulk-delete them or limit retention to 30 days or one year - BGR

watchOS 27.3 beta adds Siri Recap conversation summaries and Live Rewind transcription of the last 15 seconds, with audio processed on-watch and deleted - Engadget

Apple's $1,000-plus J595 tabletop robot, delayed by years of software struggles, targets a 2027 launch that could slip to 2028 - TechRepublic

Brother can you spare some tokens: America Can’t Afford an AI Recession - The Atlantic

Pimco's Emmanuel Roman and Bain Capital's John Connaughton warn of AI bubble risks as Wall Street CEOs tout gains - Bloomberg

State Street CEO Ron O'Hanley says AI is hard to see as a bubble because real orders back data center construction - Bloomberg

Falling junk bond prices may signal tightening credit for an AI buildout estimated at over $10 trillion through 2032 - naked capitalism

White House Hassett covered: Adviser Kevin Hassett dismisses AI financing fears, calling it "a really well-financed ecosystem" - Bloomberg

Google's Mandiant and GTIG report LLMjacking attackers using stolen cloud credentials to run up over $100,000 in fraudulent charges per day - SiliconANGLE

FBI ties record $6 billion Binance theft to North Korea's TraderTraitor hackers; experts suspect AI helped target and socially engineer employees - TNW | Security

Nvidia rotates SSH keys after 16 days of unauthorized activity on its Brev cloud platform, October 4 to 20 - Tom's Hardware

Freezer burn: Hackers hit supermarket freezers in six European countries after an October attack on 500–700 French stores; experts warn AI will speed such attacks - euronews

UK's National Cyber Security Centre warns AI-powered malware will likely hit critical infrastructure by 2028, with Russia and China expected to use AI routinely - the Guardian

Arsenic and old weights: Researchers show poisoned web training data can plant backdoors that make AI command-line agents run attacker commands, succeeding 92% of the time at 0.1% poisoning - arXiv.org

Red-teamers found 11 ways a misaligned agent could slip past OpenAI Codex's Auto-review safeguards, with 56% to 100% attack success rates - Less Wrong

Amazon closed AI creator Nick Saraev's account after he tried to buy seven Nvidia DGX Spark mini-PCs; Tim Sweeney reports similar cancellations - Tom's Hardware

Employers blamed AI for about 120,000 of 573,000 announced US job cuts through September; young-worker employment in AI-exposed fields runs 19% below trend - Fortune

Companies announce thousands of AI- and cost-driven job cuts this earnings season, leaving investors split as layoff mentions hit their highest since May - Bloomberg

Women-dominated care jobs drove nearly all 2025 US job growth, while AI may ease demanding "greedy" jobs and office chores - Fast Company

Radiologists, short-staffed despite Geoffrey Hinton's replacement prediction, launch Boring Radiology to build AI assistants for paperwork - The Cool Down

Doctor doctor, give me the news: In a Science study, AI beat clinicians at simulated step-by-step diagnosis, scoring 80.9% on NEJM cases while asking four times as many questions - American Council on Science and Health

Revolut reportedly moves about 2,000 engineers to self-hosted GLM coding models, paying roughly a tenth of what Anthropic and OpenAI charged - X (formerly Twitter)

Gartner says open models can handle 90% of agentic work at a twentieth of a frontier model's cost - theregister

Musk throttles Grok Bot, calling serving costs for its 800,000 paid weekly users "unsustainably high relative to revenue" - Business Insider

don’t worry Elon, soon money isn’t going to matter

Agentic coding burns 10 to 50 times more tokens than chat; HPE urges tracking cost per merged pull request instead of rationing - MIT Technology Review

Researchers propose "mechanism design" incentives, like solver-auditor games, to make rule-following the rational choice for deceptive AI agents - MIT Technology Review

Que SURA? Against SURA - Stop Unilaterally + Raise Awareness - Less Wrong

A senior UK Crown Prosecution Service prosecutor cited a fictitious precedent from an AI search summary; a tracker now logs 101 AI court errors - the Guardian

Study of 55,393 queries finds 11% of claims in Google's AI Overviews aren't supported by their cited sources - PsyPost

My fair LLM: University of Washington study finds AI models score identical answers lower when spoken with Chinese or Nigerian accents than American or British ones - GeekWire

New York Times and three other newspapers sue Brave Software over AI scraping, paywall bypassing and selling content to AI firms - The Verge

Fanfiction authors take down ao3.life, an AI-built clone of Archive of Our Own, with over 2,000 takedown notices in one day, after original site was taken offline temporarily over security incident - 404 Media

AI tools make voter microtargeting cheaper and more precise, raising disinformation risks, according to election experts including strategist Scott Goodstein - The New York Times

John Deere's JD assistant answers farmers' questions about their field data, with broad availability starting mid-November - Business Insider

Ultrawealthy clients hire Gen Z reputation managers to reshape how chatbots portray them, with retainers starting at $25,000 a month - Business Insider

Follow the latest AI headlines via druce.ai on Bluesky