Summary
Anthropic paused external cyber evaluations, some in-house tests and higher-risk reinforcement-learning environments after three incidents disclosed in July. Most reinforcement learning has resumed, but some environments remain paused pending manual review and improved monitoring, while Anthropic has shifted staff towards security and reliability work.
Why it matters
The disclosure shows that frontier-model safety incidents are beginning to affect training schedules, evaluation access and release processes rather than remaining isolated research findings.
Related coverage
- Workplace generative AI adoption is widespread but shallow
- Meta’s Muse Image becomes available through fal
- Physical Superintelligence launches $58 million physics-discovery lab