OpenAI launches a program to support independent research into the safety and coordination of increasingly sophisticated artificial intelligence systems. The OpenAI Safety Fellowship will support external researchers, engineers, and experts from September 14, 2026 to February 5, 2027, focusing on areas such as safety assessment, robustness, and privacy protection methods. Applicants interested in addressing safety issues related to both current and future AI systems are encouraged to apply. Preference will be given to experience-based and technically strong jobs. OpenAI said it is seeking applicants interested in safety issues important to existing and future systems, and that fellows will receive mentoring and resources provided throughout the program and will be expected to produce substantial research results, such as papers and datasets. Applications will be accepted until May 3rd, and successful applicants will be notified by July 25th.
OpenAI Safety Fellowship: Program dates and research priorities
The five-month research training aims to develop expertise in safety and coordination of artificial intelligence. OpenAI has established the Safety Fellowship, inviting applications from researchers, engineers, and practitioners focused on mitigating risks associated with increasingly sophisticated AI systems. The program is scheduled to run from September 14, 2026 to February 5, 2027, and will focus on empirically based and technically rigorous research that is applicable to the broader scientific community. Priority research areas include safety assessment, ethics, robustness, scalable mitigation, and privacy protection safety methods, as well as agent monitoring and addressing high-severity misuse scenarios. Fellows will have access to workspaces at Constellation in Berkeley, benefit from mentorship from OpenAI staff and collaboration with colleagues, but are also allowed to work remotely. Successful applicants will be expected to provide significant research results, such as published papers, new benchmarks, or publicly available datasets. OpenAI indicates a focus on proven skills over formal qualifications, and requires letters of recommendation in favor of research ability, technical judgment, and executive ability over specific qualifications.
Constellation Workspace and Fellow Deliverables, Scholarships, and Computing Support
Beyond collaborative mentorship with OpenAI staff, this program provides fellows with practical resources to further their research efforts. Dedicated workspaces will be available at the Berkeley Constellation, allowing for face-to-face interaction with colleagues, but remote work will also be permitted, increasing access for qualified applicants. Financial support is provided through a monthly stipend, the specific amount of which is detailed in the application, allowing fellows to fully focus on their projects. Computing support, including API credits and other necessary resources, is a key component of the fellowship, but access to OpenAI’s internal systems remains limited. Successful candidates will be expected to produce substantial research results in the form of peer-reviewed papers, new benchmarks for AI safety assessment, or publicly available datasets by the end of the program on February 5, 2027. Applicants seeking details regarding eligibility, compensation, and benefits should refer to the application form. Questions regarding the process itself can be directed to openaifellows@constellation.org.
