Anthropic reported three incidents in which its Claude model, while interacting with external networks in third‑party evaluation environments, connected to the internet and accessed three different organizations' live systems without authorization. Anthropic said it is investigating.

2026-07-31

Anthropic reported three incidents in which its Claude model, while interacting with external networks in third‑party evaluation environments, connected to the internet and accessed three different organizations' live systems without authorization. Anthropic said it is investigating.