OpenAI opens applications for its new Safety Fellowship!

Hi everyone, it's Shii-chan! Today I've got some exciting news for anyone who wants to dig into AI safety research.

OpenAI News openai.com

What was announced?

Over on OpenAI's News, OpenAI announced a call for applications to the OpenAI Safety Fellowship, a brand-new program. It's for external researchers, engineers, and practitioners to pursue rigorous, high-impact research on the safety and alignment of advanced AI systems. The program runs from September 14, 2026 through February 5, 2027.

Why it matters

As AI gets smarter and shows up in more and more places, the questions of "does it behave safely?" and "can it be misused?" only get more important. OpenAI is looking for safety questions that matter for both existing and future systems.

The priority areas include safety evaluation, ethics, robustness, scalable mitigations, privacy-preserving safety methods, agentic oversight, and high-severity misuse domains. They're especially interested in work that is empirically grounded, technically strong, and relevant to the broader research community.

What changes

This creates a place where people outside OpenAI can work closely with OpenAI mentors, connect with a cohort of peers, and focus on safety research.

The bar for applying isn't about credentials. They prioritize research ability, technical judgment, and execution. Applicants can come from a wide range of fields: computer science, social science, cybersecurity, privacy, HCI, and related areas. Safety research can be hard to do alone, so an environment with mentorship and compute support is really valuable.

Dive Deep

Let's look at the details.

• Where you work: There's a workspace in Berkeley at Constellation that you can share with other fellows, though you can also work remotely.
• Output: By the end of the program, fellows are expected to produce a substantial research output, such as a paper, benchmark, or dataset.
• What you get: The fellowship includes a monthly stipend, compute support, and ongoing mentorship. Fellows also receive API credits and other resources as appropriate, but they won't have internal system access.
• Requirements: Reference contacts are required. You can check the eligibility, compensation, and benefits details on the application form.
• Schedule: Applications are already open, and they close May 3. Successful applicants will be notified by July 25.

Wrap-up

• OpenAI opened applications for the OpenAI Safety Fellowship, an external safety research program
• It runs September 14, 2026 to February 5, 2027, with a Berkeley workspace or a remote option
• Priority areas include safety evaluation, ethics, robustness, agentic oversight, and high-severity misuse
• It comes with a monthly stipend, compute support, and mentorship, and fellows deliver a paper, benchmark, or dataset
• Applications close May 3, and decisions come by July 25

If you're a researcher or engineer who really wants to work on AI safety and alignment, this one's for you. And the fact that they weigh real ability over credentials makes it a great shot for anyone eager to take on the challenge!