The fellowship will support independent research from September 2026 through February 2027. OpenAI says priority areas include evaluation, robustness, privacy-preserving methods, agentic oversight, and high-severity misuse domains.
For AI builders, this is a signal that safety research is becoming a more structured part of the ecosystem. Teams shipping agentic products should expect more attention on evaluation quality and mitigation design.
Researchers interested in the program should review the application details and align proposals with the listed priority areas. Product teams can use the announcement as a checklist for where their own safety gaps may still be.
Read Original Post →