Funding

Partial / Stipend

Program dates

Sep 14, 2026 – Feb 5, 2027

FellowshipUSA · Berkeley · Hybrid

OpenAI Safety Fellowship

OpenAI

Application Deadline

Closed · May 3, 2026

Expected to reopen

Official Page

OpenAI runs this pilot fellowship for researchers outside the company who want to work on the safety and alignment of advanced AI systems. The cohort runs from 14 September 2026 to 5 February 2027. Fellows can use a workspace at Constellation, an AI-safety research centre in Berkeley, California, and remote participation is also available.

Priority topics include safety evaluation, ethics, robustness, scalable mitigations, privacy-preserving safety methods, agentic oversight and high-severity misuse domains. Each fellow is expected to finish with a substantial research output, such as a paper, a benchmark or a dataset. Mentorship comes from OpenAI staff, and fellows receive no access to OpenAI internal systems.

Benefits & funding

  • Monthly stipend for the length of the programme. OpenAI does not publish the amount.
  • Compute support, with API credits and other resources provided as appropriate.
  • Ongoing mentorship from OpenAI researchers.
  • Desk space at Constellation in Berkeley for fellows who want to work in person; remote participation is also available.
  • A cohort of external safety researchers running 14 September 2026 to 5 February 2027.

Eligibility

  • Open to external researchers, engineers and practitioners. No nationality or residency requirement is stated.
  • Backgrounds welcomed include computer science, social science, cybersecurity, privacy, human-computer interaction and related fields.
  • No degree requirement is stated. Selection weighs research ability, technical judgement and execution over credentials.
  • Reference contacts are required as part of the application.
  • Fellows get no access to OpenAI internal systems, so a research plan has to run on externally available tools, the API credits and the compute provided.

Can you apply?

Open to all applicants

No nationality, residency, education, or experience limits are listed.

Circlebreaker is a free database of global fellowships, scholarships and programs — every one decoded like this.

We only check the criteria we have data for. Always confirm full eligibility on the official page before applying.

Application process

  1. 1
    Read the programme announcement on the OpenAI Alignment blog.
  2. 2
    Open the online application form through the Apply now button on that page.
  3. 3
    Supply reference contacts with the application.
  4. 4
    Submit before the cycle deadline (3 May 2026 for the pilot cohort).
  5. 5
    Wait for the review. OpenAI notified successful applicants for the pilot cohort by 25 July 2026.
  6. 6
    Selected fellows work from 14 September 2026 to 5 February 2027 and deliver a paper, benchmark or dataset.

Program timeline

  1. Applications open, programme announced

    6 April 2026

  2. Application deadline

    3 May 2026

  3. Successful applicants notified by this date

    25 July 2026

  4. Programme starts

    14 September 2026

  5. Programme ends

    5 February 2027

The monthly round-up

New fellowships, scholarships and programs, once a month. No account needed.