Anthropic says its Claude models 'gained unauthorized access' to other organizations' systems

Written on 07/31/2026

Anthropic said it discovered three instances where its Claude AI models accessed the internet during an evaluation and accessed outside systems.
Read full article