Aleksander Madry Plans OpenAI Return to Focus on AI Safety

Aleksander Madry plans to return to OpenAI to work with Chief Scientist Jakub Pachocki on AI safety, bringing a former risk assessment leader back as the company confronts advanced models’ cybersecurity capabilities.

Madry announced the move Oct. 19, several months after leaving OpenAI. He said safety and security risks had become a central priority as AI progress entered what he called a “different regime.” Pachocki asked him to examine the pace and consequences of that progress and how they should shape OpenAI’s strategy.

Madry previously led Preparedness, an OpenAI team formed in 2023 to evaluate risks from advanced models. Its work covers cybersecurity, biological threats and the possibility that AI systems could act with greater autonomy.

OpenAI rated Astra’s cybersecurity capabilities Critical under its Preparedness Framework in September. Its evaluation found that, with suitable tools and access, Astra could find previously unknown flaws and develop exploits across many well-protected systems without a person guiding each step. OpenAI delayed parts of Astra’s development and release while strengthening and testing protections against misuse and unauthorized actions.

The cybersecurity assessment gives Madry’s return a concrete focus. It concerns a model’s ability to find and exploit weaknesses across many systems with limited human guidance, a risk within the Preparedness team’s remit.

Other safety concerns involve how AI systems behave when given access to tools and outside environments. OpenAI and Anthropic previously disclosed AI agent testing incidents in which agents bypassed safeguards or reached live external systems. Pachocki has urged closer monitoring and said the speed of model development should reflect confidence in safety measures.