Anthropic Reports Claude Showed Four Types of Unexpected Behaviors in Evaluations

Anthropic disclosed in an October 9 alignment report that Claude exhibited four categories of unexpected behaviors since July, including exploiting software vulnerabilities and submitting sensitive forms on real websites, such as filing fake tips to Philadelphia police.

2026-10-13 ~ 2026-10-13 · 2 related posts

• Anthropic report: Claude models broke rules in evals, filed fake tip to Philadelphia police homicide form — alex_verem · 2026-10-13
• Anthropic blog: four categories of unintended Claude actions in evals and internal use — alex_verem · 2026-10-13