TL;DR

OpenAI has opened applications for a new Safety Fellowship aimed at external researchers, engineers and practitioners working on the safety and alignment of advanced AI systems. The pilot runs from 14 September 2026 to 5 February 2027, with applications closing 3 May and successful applicants notified by 25 July. Fellows receive a monthly stipend, compute support, and the option to work from Constellation’s Berkeley workspace.

What OpenAI Is Funding

The programme’s published priority areas read as a snapshot of the lab’s current safety preoccupations: safety evaluation, ethics, robustness, scalable mitigations, privacy-preserving safety methods, agentic oversight, and high-severity misuse domains. OpenAI says it wants empirically grounded, technically strong work “relevant to the broader research community” — a deliberate signal that this is not a closed in-house alignment programme.

Fellows are expected to produce a substantial research output by the end of the fellowship, such as a paper, benchmark or dataset. They get API credits and other resources, but explicitly will not have internal system access — a notable framing that positions the programme as external research support rather than embedded employment.

Where It Sits in OpenAI’s Safety Posture

The fellowship lands alongside OpenAI’s recent moves to expand its external safety footprint, including its bug bounty programme launched in March and updated teen safety policies for GPT-OSS Safeguard. It also follows the broader trend across frontier labs — Anthropic and Google’s DeepMind run analogous external programmes — of trying to grow the alignment researcher pipeline without scaling internal headcount linearly.

Looking Forward

For UK-based AI safety researchers and PhD students, the practical implication is another funded route into frontier safety work that does not require relocating into a permanent corporate role. The Berkeley workspace is optional, and the May application deadline gives UK academics time to pull together a competitive submission. The wider question is whether external fellowships of this kind genuinely shape lab decisions or whether they function mainly as community signalling — something that will only become clear once the first cohort’s outputs are published.