
Anthropic releases Claude Fable 5, a public Mythos-class model with guardrails that route cybersecurity, biology, and chemistry queries to Opus 4.8 - The New York Times
Anthropic now retains Mythos traffic for 30 days, overriding previous zero-retention enterprise agreements - TechCrunch
don’t use Fable to make OpenAI IPO pitch decks
Fable 5 scores 80.3% on SWE-bench Pro, topping GPT-5.5; Stripe completed a 50M-line Ruby migration in one day - VentureBeat
Anthropic's Fable 5 uses broad safety classifiers that block benign cancer and cybersecurity questions, reverting to Opus 4.8. - Business Insider
Ethan Mollick on Fable/Mythos - One Useful Thing
Claude Fable 5 - Simon Willison
SpaceX unveiled Gigasat, an 11-million-square-foot factory to manufacture AI1 orbital data-center satellites targeting 1 GW of space AI compute by 2027 - Tom's Hardware

X's automated systems told a Chinese dissident in the UK that deepfake sexual abuse posts targeting her did not breach platform rules. - the Guardian
SpaceX investors bullish on CSAM
Crusoe holds 5GW in data center contracts but paused a 1.8GW Wyoming campus at a customer's request - Bloomberg

GPU idle rates exceed 30% across hyperscaler fleets, costing trillions; a ByteDance engineer's OpenSkill automation cut idle time 35% in production - TNW | Artificial-Intelligence
White House orders federal AI testing unit to halt public model assessments, citing national security concerns - The Wall Street Journal




Microsoft's June 2026 Patch Tuesday fixed a record ~200 Windows flaws, with AI tools increasingly used to discover vulnerabilities. - Krebs on Security

Would-be Walter Whites use AI to synthesize illicit drug precursors and evade restrictions - Financial Times

Elon Musk claims Tesla's AI5 chip will outperform Nvidia's best by 2-3x at one-tenth the cost, with production targeting 2027 - 24/7 Wall St.
bit of a gap between promises and execution


MIT's ultrasound wristband tracks 22 degrees of hand motion for real-time robot teleoperation at 120ms, no cameras or gloves needed - TNW | Deep-Tech
PR-CAD unifies text-to-CAD generation and editing into one agent trained with reinforcement learning, achieving top scores on controllability and faithfulness - arXiv.org
Administrative workers face greater AI displacement risk than coders and are least able to adapt, Northwestern economists find - The New York Times
How 16 top economists think AI will change the job market, and how to prepare - The Wall Street Journal
Google DeepMind economist finds no AI jobs bloodbath yet, but warns of a cascade of investor-signaling layoffs - Business Insider
McKinsey now runs 25,000 AI agents alongside 40,000 consultants, compressing 'week-one answers' to one hour - Business Insider

Chief Agent Provocateur: Satya Nadella argues AI agents deserve real identities like employees, noting he personally manages 100 coding agents - Windows Central

AI proves coding was never the main job — developers must now specify, validate, and verify rather than implement - Hacker Noon

Developer uses AI agents to build Grit, a Rust reimplementation of Git that passes over 99% of the Git test suite. - Butler's Log
LLMs are advising real military targeting decisions, MIT Technology Review finds - MIT Technology Review
Geoffrey Hinton says Ukraine's drone warfare has complicated his earlier push for a ban on lethal autonomous weapons - Business Insider
maybe OK for defense?
The bitter lesson, part deux: AI pioneer Rich Sutton argues generative AI trained by supervised learning cannot achieve true creativity or scientific discovery without an evaluation mechanism. - X (formerly Twitter)

Follow the latest AI headlines via druce.ai on Bluesky

