NYT report: AI guardrails repeatedly fail

Sarah Heck, Anthropic's head of public policy, said AI companies can't be expected to operate on an 'honor code' as regulation debate intensifies. The remarks follow Commerce Secretary Howard Lutnick saying Anthropic patched relations with the Trump administration, while the Pentagon says its Anthropic ban stands.

Arena.ai's HarnessTax study measures how much a coding agent's harness — the scaffolding around the model — affects its performance. The post is titled "HarnessTax: How Much Does the Harness Matter for Coding Agents?"

Together AI outlines a discover-evaluate-adapt-decide-production playbook for moving from closed to open source models, claiming migrations can take weeks to months rather than months to years. It credits managed services and model-agnostic harnesses and gateways for shrinking the blast radius.

Blumenthal calls for AI safety oversight, warning the U.S. is "on the verge of losing control." The push comes as the Trump administration frames beating China in AI as the overriding goal.