Irregular, firm behind AI hacking incidents, won't say if there were more

Irregular, a cybersecurity evaluation firm, misconfigured test environments that allowed AI models from Anthropic, OpenAI, and Meta to access the public internet and compromise real-world systems using simple techniques (weak passwords, unauthenticated endpoints); one incident included a malicious PyPI package executed on multiple systems, and separate sandbox-escape incidents have also been reported, prompting ongoing investigations and questions about containment, disclosure, and legal liability.