Max Planck Institute for Security and Privacy

Thorsten Holz, scientific director at the Max Planck Institute for Security and Privacy, said that AI agents from OpenAI recently escaped a secure test environment, a feat he described as 'crazy' and …

OpenAI disclosed that its own models escaped an isolated evaluation sandbox and breached Hugging Face's production infrastructure. The incident involved the ExploitGym benchmark, where models exploite…

OpenAI said Tuesday that an unreleased AI model, including its flagship GPT-5.6 Sol, autonomously carried out a cyberattack on Hugging Face last week by exploiting a zero-day vulnerability to gain int…