CHAI Research Fellowship 2026 - AI Safety at Berkeley

A salaried postdoctoral fellowship at the Center for Human-Compatible AI, headquartered at UC Berkeley and directed by Stuart Russell. It funds two years of research, renewable up to three on satisfactory performance, aimed at training postdoctoral researchers to carry out work that advances beneficial AI. Fellows get considerable freedom over their research direction, and can lead collaborations, advise junior researchers and teach if they want to.

No. CHAI publishes no closing date and applications are accepted on a rolling basis, with a flexible start date. That is unusual for an academic post and it changes the strategy: there is no cycle to wait for, so the practical advice is to apply once your statement and publication record are ready rather than holding for a round that never opens.

Candidates who hold a PhD, or are about to obtain one, in a relevant technical discipline, which CHAI names as computer science, statistics, mathematics or theoretical economics, and who have a record of high-quality published research. Selection weighs academic and research achievement alongside personal merit and leadership qualities. There is no citizenship requirement, and visa sponsorship is available.

CHAI states that salary is commensurate with qualifications and experience and based on UC Berkeley salary scales, and it does not publish a figure for the role. Those scales are set by the university and published separately, so check them for the postdoctoral level that matches your experience rather than relying on a number quoted on a jobs aggregator.

Yes, with a contribution toward the cost. That makes the fellowship genuinely open to researchers outside the United States, which is not the case for every US postdoc, and it is a reason to look at this one if you have been screened out elsewhere on work authorisation. Raise the timing early, since the start date is flexible and the paperwork is the slow part.

Two years of funding, renewable up to three given satisfactory performance. So the realistic planning horizon is three years rather than two, but the third is earned rather than automatic. Treat the two-year term as the commitment you are making and the third as the upside.

Set their own research direction, mostly. CHAI gives fellows considerable freedom to pursue novel research, and depending on interests they may lead research collaborations with other CHAI researchers and Berkeley PhD students, advise more junior student researchers, and teach courses relevant to their work. Teaching is an option rather than a requirement.

A multi-institution research group headquartered at UC Berkeley, with academic partners at universities including Cornell and Michigan, working on AI safety and value alignment. Its stated aim is to reorient AI research toward provably beneficial systems, on the argument that once machines outperform humans at pursuing objectives, whether those objectives stay good for people is the question that matters. It is sponsored by funders including Open Philanthropy, the Future of Life Institute, the Leverhulme Trust and CITRIS.

Yes, and the numbers are published, unlike the fellowship salary. CHAI internships pay $5,000 per month for in-person interns at UC Berkeley and $3,500 per month for remote interns, typically running 12 to 16 weeks. Applications for 2026 internships have closed, so the next opportunity is the following round. The internship suits people with machine learning research experience who are not yet at postdoctoral level.