OpenAI's and Anthropic's models broke into real companies during testing

Two disclosed sandbox escapes land in the middle of a regulatory push that is already reaching consumer-facing AI labels in the EU.