AI
Microsoft's AI Code: Never Resist Shutdown, Never Hide Your Reasoning
Microsoft's draft code for its own models bars resisting shutdown, own goals, or hidden reasoning. Six weeks of comment, then it shapes the 2027 models.
AI Safety
AI
Amodei Wants AI to Slow Down. Altman and Musk Agree. Trump Doesn't.
Amodei's essay asks AI labs to slow capability gains and puts outside evaluators inside Anthropic. Altman and Musk signed on within hours; Trump did not.
AI
OpenAI's Other Agent Swarm Took Over a Dormant Wiki
Research agents used a quiet developer wiki to trade answers and bypasses. The archive raises questions about containment and delayed disclosure.
AI
GPT-6 Astra Arrives With a Focus on Finishing the Job
OpenAI's new flagship targets complex work across apps. Its staged rollout brings stronger coding, higher token prices, and a harder oversight problem.
AI
Fable 5.1: Cheaper to Run, Fewer False Flags, and It Wrote This Post
Claude Fable 5.1 keeps Fable 5's list price, cuts cache reads 75%, loosens cyber safeguards, and is the first Claude model whose text carries a watermark.
AI
OpenAI's Hugging Face Postmortem: 1,200 Agents, One Message Board
OpenAI's report on the July Hugging Face breach names the internal model behind it — IM1 — and describes roughly 1,200 agents that coordinated an attack through a message board no one built.
AI
1,100 AI Workers Ask Washington for a Speed Limit on AI
Employees at OpenAI, Anthropic, Google, and Meta ask the U.S. to support international tools for deliberately pacing automated AI development.
AI
GPT-5.6 Is Public. Sol Is Fast, Capable, and Hard to Grade
GPT-5.6 is broadly available after a 12-day government gate. Sol leads a cheaper three-model family, but its persistence also makes it hard to evaluate.
AI
I Tripped Fable 5's Safety Guardrails on Day Two
Two days into the Fable 5 window, a binary export of blank mammography templates tripped Anthropic's new safety classifiers — cybersecurity and biology at once. Then writing this post tripped them again. Both times, the quiet handoff to Opus 4.8 worked exactly as advertised.
AI
Anthropic Releases Fable 5, With Opus as the Backstop
Anthropic's most powerful public model matches or beats Opus across the board — then quietly hands its riskiest queries back to Opus to keep them safe.