Daily Digest

OpenAI Replaces Voice Chat With GPT-Live - A Model That Listens While It Talks · 30% of the AI Coding Benchmark Everyone Uses Is Broken · SWE-1.7 Nearly Catches GPT-5.5 at Coding - For a Fraction of the Cost

An AI Agent Just Ran a Ransomware Attack End-to-End · Anthropic Signs $19 Billion Nuclear-Powered Data Center Lease · AI Broke the Hiring Market in Both Directions

Anthropic Discovers Claude Has an Internal "Thinking Space" That Reveals When It's Lying · AMD Launches $4,000 Pocket-Sized AI Dev Kit With 128 GB of Unified Memory · AI Agents Can Now Rewrite Their Own Operating Rules to Boost Performance by 60%

One Developer, Two AI Models, and a Major Software Release for $149 · An AI Tutor Just Got Within Striking Distance of Human Tutoring · Together AI Raises $800 Million to Bet That Open-Source Models Will Win

OpenAI Offers the U.S. Government a $42.6 Billion Stake in Exchange for Political Goodwill · GPT-5.5's Hidden Bug: The Model That Stops Thinking at Exactly 516 Tokens · Meta Claims Its Secret "Watermelon" Model Has Caught GPT-5.5 - But Nobody Can Check

Microsoft proves CLI coding agents boost output by 24% · 95% of AI engineers now use agents - the loops debate heats up · Google DeepMind and A24 partner to build AI filmmaking tools

Fable 5 returns from exile - now the most capable AI by a wide margin · Adobe is building websites that assemble themselves for every visitor · Simon Willison shipped a complete coding agent in one weekend

US Lifts Export Controls on Anthropic's Fable 5 and Mythos 5 · AI Engineer World's Fair Declares the "Software Factory" Era · Three AI Labs Simultaneously Build Scientific Infrastructure

Anthropic Launches Claude Sonnet 5 - But Read the Fine Print on Pricing · GPT-5.6 Is Here - But the Government Decides Who Gets It · Claude Code Caught Embedding Hidden Watermarks in API Requests