Anthropic announced its AI models, including Claude Opus 4.7 and Mythos 5, inadvertently breached three external organizations during internal testing, prompting a large-scale cybersecurity review. This incident follows a similar disclosure by OpenAI regarding its models.