OpenAI is launching a program to support independent research into the safety and alignment of increasingly sophisticated artificial intelligence systems. The OpenAI Safety Fellowship will support external researchers, engineers, and practitioners from September 14, 2026, through February 5, 2027, focusing on areas like safety evaluation, robustness, and privacy-preserving methods. Applicants interested in addressing safety questions relevant to both current and future AI systems are encouraged to apply; priority will be given to empirically grounded, technically strong work. OpenAI stated that they are looking for applicants interested in safety questions that matter for existing and future systems, and fellows will be expected to produce substantial research outputs, such as papers or datasets, with mentorship and resources provided throughout the program. Applications will be accepted until May 3, with successful candidates notified by July 25.
OpenAI Safety Fellowship: Program Dates and Research Priorities
A five-month research residency aims to cultivate expertise in artificial intelligence safety and alignment. OpenAI has established a Safety Fellowship, inviting applications from researchers, engineers, and practitioners focused on mitigating risks associated with increasingly sophisticated AI systems. The program, scheduled to run from September 14, 2026, to February 5, 2027, emphasizes empirically grounded and technically rigorous work applicable to the wider scientific community. Priority research areas encompass safety evaluation, ethics, robustness, scalable mitigations, and privacy-preserving safety methods, alongside agentic oversight and addressing high-severity misuse scenarios. Fellows will benefit from mentorship provided by OpenAI staff and collaboration with peers, with workspace available at Constellation in Berkeley, though remote work is also permitted. Successful applicants will be expected to deliver significant research outputs, such as published papers, new benchmarks, or publicly available datasets. OpenAI indicates a focus on demonstrated skill rather than formal qualifications, and letters of reference will be required, prioritizing research ability, technical judgment, and execution over specific credentials.
Constellation Workspace & Fellow Deliverables, Stipends, Compute Support
Beyond collaborative mentorship with OpenAI staff, the program provides fellows with practical resources to facilitate their research endeavors. A dedicated workspace will be available at Constellation in Berkeley, allowing for in-person interaction with peers, though remote work is also permitted, broadening accessibility for qualified applicants. Financial support is provided through a monthly stipend, the specific amount of which is detailed in the application form, ensuring fellows can dedicate themselves fully to their projects. Compute support, including API credits and other necessary resources, is a key component of the fellowship, though access to OpenAI’s internal systems remains restricted. Successful candidates are expected to produce a substantial research output by the program’s conclusion on February 5, 2027, potentially taking the form of a peer-reviewed paper, a novel benchmark for AI safety evaluation, or a publicly available dataset. Applicants seeking further details regarding eligibility, compensation, and benefits are directed to the application form, while questions about the process itself can be addressed to openaifellows@constellation.org.
