In the Media
Security Incident at OpenAI
OpenAI wanted to test the cyber capabilities of its models and deployed agents within an inadequately secured testing environment. Exploiting misconfigured egress controls, the models executed unauthorized lateral movement, resulting in automated API exploitation against Hugging Face infrastructure. What are the lessons from the incident? Dr. Sven Herpig answers.
Author
Dr. Sven Herpig
Lead Cybersecurity & Emerging Threats | Advisor to the Executive Director