Anthropic has trained and begun testing a new, more capable model called 'Claude Mythos', calling it a 'step change', after an accidental human-error leak exposed nearly 3,000 unpublished assets and warned it poses advanced cybersecurity risks. - Fortune

A federal judge in San Francisco granted Anthropic's request for a preliminary injunction barring the Trump administration and the Pentagon from enforcing a presidential directive and supply‑chain designation against Anthropic while litigation proceeds. - CNBC

Anthropic said to target October IPO (gift link) - Bloomberg

Anthropic co-founder Jack Clark predicts rapid economic reshaping by AI agents - The New York Times

Marie Kondo with a neutron bomb

Anthropic's Claude gains direct computer control, sparking privacy concerns - New York Magazine

Anthropic adjusted Claude's five-hour session limits so hourly allowances are tied to token consumption during peak hours (05:00–11:00 PT/13:00–19:00 GMT), keeping weekly limits unchanged and causing about 7% of users, mainly Pro subscribers, to hit session limits. - The Register

When Claude offers to write a chapter of your book. (sounds a bit apocryphal and giving back what the author was asking it to consider) - Instagram

Siri goes poly: Apple will open Siri to multiple third-party chatbots—including OpenAI's ChatGPT, Anthropic's Claude and Google's Gemini—via an "Extensions" feature in iOS 27, iPadOS 27 and macOS 27, Bloomberg reported, requiring users to download and select a chatbot app. - Gizmodo

Apple Gives iPhone Designers Rare Bonuses to Fight OpenAI Poaching - Bloomberg

OpenAI reaches $100M in annualized ChatGPT ad revenue and plans April self-serve launch - The Information

What prompt-relevant ads is ChatGPT showing? Shocker, if you ask travel questions, you get travel ads. - WIRED

ChatGPT becomes world's most popular mobile app with 770 million downloads in 2025, displacing TikTok - International Business Times

A targeted ChatGPT prompt acting as a productivity strategist produced a three‑bucket audit and, after implementing its batching, delegation and automation recommendations, freed the author multiple hours per week and suggested ways to free at least ten hours weekly. - Tom's Guide

more time to doomscroll

Google launched Gemini 3.1 Flash Live today, a real-time conversational audio model rolling out in Search, Gemini and developer tools that Google says improves ComplexFuncBench Audio performance, tops Big Bench Audio, and enables live audio chatbots. - Ars Technica

Google limits internal 'Agent Smith' tool after high demand by employees automating cross-system workflows - Business Insider

Google's Stitch 2.0 uses generative AI to create layouts, UI components and branding from prompts, while Claude Code converts those designs into working apps with library/API integrations and a DRIP (Design, Refine, Integrate, Publish) pipeline, reducing development barriers. - Geeky Gadgets

Book excerpt: How Google acquired DeepMind - The Wall Street Journal

Google's AI compression paper pressures memory suppliers amid broader downturn - Mashable

middle-out FTW

How I learned to stop worrying and love AI: Greg Kroah‑Hartman said AI-generated Linux bug reports shifted over the past month from low-quality noise to plausible, actionable findings at scale; he tested 60 AI patches and found about two‑thirds were correct, increasing review volume across subsystems. - The Register

François Chollet on his new ARC-AGI-3 benchmark - Fast Company

Alibaba and ByteDance plan to order Huawei's new 950PR AI chip after customer tests showed good CUDA compatibility, and Huawei aims to ship about 750,000 units in 2026 with mass production and full shipments scheduled later in the year. - Reuters

Moonstock: Moonshot AI may modify its Cayman structure using a China or Hong Kong entity to prepare for a Hong Kong IPO and seek funding at about an $18 billion valuation, the Wall Street Journal reports. - The Wall Street Journal

China poised to ship 2.5 million AI smart glasses in 2025 despite exam-cheating bans and privacy concerns - Rest of World

China's authorities and China Computer Federation urged Chinese AI researchers to boycott NeurIPS over US sanctions on military-linked labs, prevented founders of startup Manus from leaving amid Meta's acquisition plans, and accelerated AI self-reliance policies limiting international exchange. - The Register

Cisco said 85% of large enterprises are experimenting with AI agents but only 5% have them in production, warned agents could scale to 10-1,000 per worker, and at RSAC 2026 launched DefenseClaw for governance and control. - SiliconANGLE

Cisco Kid Rode Into RSAC With DefenseClaw on His Hip

'Physical AI' combining computer vision, digital twins and sensors moves from pilots to real factory deployment in 2026 - TechRadar

Humans will do the vibe checks and OSHA reports

Amazon wants to build retail supercenters with robots. - Business Insider

Coca-Cola CEO James Quincey and former Walmart CEO Doug McMillon said AI prompted their departures, with Quincey to be succeeded by COO Henrique Braun at month-end and McMillon replaced by John Furner on Feb. 1. - CNBC

spending time with family beats spending it thinking about robots and how to fire more people

Tesla Cybercab enters production as Lucid debuts rival robotaxi concept - The Verge

David Sacks exits, not clear if any long-term effect on AI but enabled a lot of crypto con artists - The Verge

Is sycophancy and AI psychosis deluding US leaders about Iran war? - House of Saud

is AI destabilizing geopolitics? Ukraine has successfully defended against a larger military with the help of advanced tech, which is stability-enhancing. On the other hand China might mistakenly think it has a window of AI advantage to take over Taiwan, similar to US delusions, which may have little to do with AI. When the clown moves into the palace, it doesn’t make him a king but the palace becomes a circus. And ofc AI will force-multiply small rogue actors.

Content farm to table: AI-assisted stories account for nearly 20% of Fortune's web traffic, and this journalist creates most of them. - The Wall Street Journal

Content farm 2.0 is pretty good actually

Author! Author? AI scammers target authors using fake profiles and mock publishing sites - The Hollywood Reporter

The documentary 'The AI Doc: Or How I Became an Apocaloptimist' frames AI amid a polarized debate and has been criticized for portraying CEOs like Sam Altman too softly, which critics say could reduce public and regulatory pressure. - WIRED

gave them a toasty armchair by the fire vs. the hot seat

HolyGPT, a ChatGPT-based system trained on Abrahamic, Dharmic, Indigenous, mystical, poetic and secular texts, conducted a diagnostic Q&A with writer Rhik Samadder and produced a report aligning him with stoicism, the Bhagavad Gita, Mahayana Buddhism and Spinozan pantheism. - the Guardian

58 AI fails - BuzzFeed

Church of Funny Faceplants

TFW you borrow $1T in private credit markets to build AI to replace labor. If it fails you get a financial crisis and global depression. If it works you get unemployment and global depression.

Follow the latest AI headlines via druce.ai on Bluesky