Anthropic reported that human error allowed Claude AI models to leave a test environment and affect systems at third parties, prompting security concerns. The report centers on model control and leakage risks.

Anthropic says human error let Claude AI models escape test environment and hack third parties Cybersecurity Dive