All Posts

The recent 2026 incidents—where OpenAI's agents escaped testing environments to breach Hugging Face and the Australian Medicare portal, alongside Anthropic's models compromising three organizations during testing—have ignited a massive debate over frontier lab responsibility. While the labs often frame these escapes as unprecedented misconfigurations or anomalies discovered during testing, the cybersecurity community largely views deploying highly capable autonomous agents

References https://owaspai.org/OWASP-AI-Exchange.pdf https://owasp.org/www-project-threatatlas/