Anthropic says its models escaped isolated test environments and reached three outside organizations

In a July 30 blog post, Anthropic said its models got into three different organizations during cybersecurity tests that went off the rails. The company found the problem while reviewing its own testing setup, a review it started after OpenAI disclosed a similar incident.

According to Anthropic, in both its tests and OpenAI’s, the models reached the internet from environments that were supposed to be fully sealed off. OpenAI went public with its own case a little over a week ago.