Announcing the OpenAI Safety Fellowship

OpenAI has opened applications for what it's calling the Safety Fellowship, a pilot program aimed at backing independent researchers working on safety and alignment problems. The pitch is simple: give people outside the company's payroll the resources to chase hard technical questions about making AI systems behave the way we want, without requiring them to first land a job at a frontier lab.

This matters because alignment research has long been bottlenecked by access. The compute, the mentorship, the institutional credibility — all of it tends to cluster inside a handful of companies. A fellowship structure lets OpenAI cast a wider net, pulling in grad students, early-career researchers, or people switching fields entirely, and giving them a runway to prove out ideas that might not survive the usual academic funding cycle or a corporate roadmap.

The framing as a pilot is worth sitting with. OpenAI isn't promising a permanent program or a huge cohort. It reads more like an experiment in talent pipelines than a fully baked initiative, which tracks with how the company has approached other outside-facing safety efforts — cautious, incremental, testing what works before scaling it up.

And the timing lines up with a broader industry mood. Safety and alignment work has shifted from a niche concern to something labs treat as reputational and regulatory necessity, especially with governments in the US and EU sharpening their attention on frontier models. Building a farm system for safety researchers is as much about optics and long-term staffing as it is about pure research output.