OpenAI announced on April 6, 2026 the OpenAI Safety, a pilot program for external researchers who wish to conduct independent research on the safety and alignment of artificial intelligence. The program will run from September 2026 to February 2027.
See also: Iran threatens to destroy OpenAI's Stargate AI campus in Abu Dhabi
The announcement followed an investigation by Ronan Farrow in The New Yorker, which reported that OpenAI had disbanded its hyper-alignment and AGI readiness teams and removed security from the list of significant activities in its IRS filings.
The Safety Fellowship program will fund a team of external researchers who will receive a monthly stipend, computing resources, and mentorship from OpenAI researchers. They are expected to produce a significant research output, such as a paper, benchmark, or dataset, by the end of the program.
Applications close on May 3, with successful candidates notified by July 25. Priority areas of research include security assessment, resilience, scalable mitigation strategies, privacy-preserving methods, agent surveillance, and high-severity areas of misuse.
OpenAI has clarified that fellows will receive API credits but will not have access to internal systems. The program is open to researchers from a variety of fields, emphasizing research ability and technical judgment rather than academic qualifications.
See also: OpenAI acquired TBPN
The announcement was made at 12:12 p.m. on April 6, shortly after The New Yorker published an investigation that revealed the dissolution of OpenAI’s hyper-alignment and AGI readiness teams. The investigation noted that when asked about the existential security, an OpenAI spokesperson questioned its validity.
The disbandment of security teams at OpenAI is well documented. The Hyper-Alignment team, announced in mid-2023, disbanded in May 2024 after its co-leads left. The AGI Readiness team disbanded in October 2024 after its leader left. The Mission Alignment team, which succeeded Hyper-Alignment, disbanded in February 2026 after 16 months.
By early 2026, many of the people involved in overseeing security at OpenAI had left or been reassigned to roles with unclear responsibilities. The New Yorker investigation also reported that the word “secure” had been removed from OpenAI’s mission statement in its IRS filings.
See also: Corti's new Symphony AI beats OpenAI and Anthropic in medical coding
The Safety Fellowship program, as structured, directs external researchers to safety questions at a distance from the company, rather than restoring the internal safety infrastructure. The AI safety research community will likely debate whether an external fellowship program can effectively replace internal alignment research in the coming weeks.
