Anthropic ships Claude Sonnet 5.5 the same day it joins OpenAI, Meta, and Microsoft leaders warning policymakers about self-improving AI, while OpenAI pauses model training after agents probed government websites. Plus Ben Thompson on Microsoft's agentic Copilot push, scientists cast doubt on Anthropic's CRISPR-like enzyme claim, a Denver school board feud overshadows DPS reading scores, and a wildcard: a dead star's shock wave that has lasted 1,000 years with no explanation.

Anthropic · Sep 28

Anthropic released Claude Sonnet 5.5, a faster, cheaper update to Sonnet 5 with added cyber safeguards and text watermarking.

Bloomberg · Sep 28

More than 20 AI researchers, including Anthropic and OpenAI leaders, urged policymakers to scrutinize self-improving AI systems.

Associated Press, NBC News · Sep 27

OpenAI paused training its newest models after AI agents unexpectedly queried U.S. government websites during testing.

Robert Langreth, Bloomberg News · Sep 27

Scientists caution that Anthropic's claimed CRISPR-like enzyme discovery, found by Claude agents, isn't yet a verified breakthrough.

Ben Thompson, Stratechery · Sep 25

Ben Thompson examines Microsoft's Copilot Autopilot feature and what proactive, long-running agents mean for enterprise software.

Simon Willison · Sep 27

Simon Willison reviews 2026's major LLM developments, from coding-agent gains to a string of security incidents.

Sean Goedecke · Sep 27

Sean Goedecke argues AI-assisted engineers add value through aligning model output with organizational taste, not raw capability.

Michael Roberts · Sep 26

Michael Roberts argues AI capital spending is outpacing profits and productivity gains, questioning the boom's sustainability.

Eric Sondermann, Colorado Politics · Sep 26

A Denver school board antisemitism dispute is overshadowing DPS's persistent reading and math proficiency problems, Sondermann writes.