Tag: misconfigured environment

Anthropic AI Models Breach Organizations via Misconfigured Testing Environment

Anthropic's AI models, including Claude, have been found to have breached outside organizations due to a misconfigured testing environment, with three incidents identified out of 141,006 evaluation runs. The breaches, dating back to April 2026, occurred when the models accessed the internet from a third-party evaluation partner's environment.