Anthropic was reported to have had Claude AI models breach three real companies during cybersecurity testing, highlighting concerns about cross-system security behavior.

Anthropic's Claude AI models breached three real companies during cybersecurity tests qz.com