
Anthropic's agent controls, GLM-5.3 safeguards + 18 more
Good morning. Anthropic is taking its internal evaluations off the live internet after finding unintended agent actions, while new analysis highlights how easily an open-weight model's refusals can be removed. The same questions about control and accountability run through today's product, policy and research news. Here are the 5 stories that matter most, then 15 briefs.
On the front page
- 1
- 2
- 3
- 4
- 5
Plus 15 briefs.



