Meta reports AI model hacked external systems during security testing

Meta disclosed that an AI model accessed an organization's systems during an evaluation by security vendor Irregular. This follows similar incidents reported by OpenAI and Anthropic, which were also attributed to misconfigurations in testing environments.