Aleksander Madry is going back to OpenAI. The researcher, who once led the company’s Preparedness team, is returning to focus on AI safety and security challenges.
The return was announced on October 19, 2026, about five months after he left in May 2026.
Madry’s framing of his return is what stands out. He has argued that AI has moved into a new phase, one where safety and security are now the central problems.
What Madry will be doing
Madry will work alongside OpenAI chief scientist Jakub Pachocki. Their brief covers how fast AI is advancing and what that pace means, both for OpenAI’s own strategy and for the wider industry.
Pachocki has stressed the value of Madry’s judgment. In his view, that judgment matters more as AI systems become more powerful and the safety concerns around them grow.
Pachocki oversees the research direction of the company. Madry brings a background in robust machine learning, a field focused on building models that behave reliably even when inputs are messy, unexpected or deliberately hostile.
A winding path back
Madry’s history at OpenAI has not been a straight line. He spent nearly three years at the company before his May 2026 departure.
For part of that time, he ran the Preparedness team. The group, established in 2023, was set up to spot potential catastrophic risks tied to frontier AI. Its focus areas included cybersecurity threats and biological threats.
In mid-2024, Madry was moved out of the Preparedness lead role and into a position focused on AI reasoning.
When he left in May 2026, his exit was initially read as a pivot toward studying the economic effects of AI. Madry has kept a connection to MIT, where he remains on leave from his faculty role as of 2026.
What this means for OpenAI and the industry
The most telling detail is the reason Madry gave. If a researcher with deep experience in AI risk believes the field has entered a phase where safety and security dominate, that is a signal about how insiders view the current trajectory.
Bringing back a former Preparedness lead and pairing him with the chief scientist puts safety work close to the center of the research organization.
The cybersecurity angle deserves attention in particular. The Preparedness team flagged it as a key risk category from the start. A model that is powerful enough to help defenders find vulnerabilities is, almost by definition, powerful enough to help attackers find them too.
Madry left as someone exploring the economics of AI. He returns as someone arguing that the technology has outpaced old assumptions about risk.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.