Generator Residency

Regular applications close April 27, 2026. If you need an earlier answer, you can opt into early decision on the application form; early decision applications close April 17, 2026. Click here to apply.

Early decision applicants will hear back by around May 1, 2026. Regular applicants will hear back by around May 10, 2026.

If you have a hard deadline to commit somewhere else, like another program, a job offer, or a visa timeline, you can opt into early decision on the application form. Early decision applicants need to apply by April 17, 2026 and will hear back by around May 1, 2026. Everyone else goes through regular decision (apply by April 27, hear back by around May 10). Opting into early decision doesn't affect your chances; it just moves your timeline up.

You will submit your initial application. If invited to continue, you will complete a ~2 hour trial task followed by a short (~15 minute) interview, after which we will make final acceptance decisions.

June 15 – August 28, 2026 for the core residency. The first two weeks match residents to projects and help them build context on the field. The remaining nine weeks are project execution. Selected residents are then invited into a three-month extension period, roughly September – November 2026.

We help residents land a full-time role at an AI safety org, spin their project into a new org, or hand it off to one that can keep it going. Selected residents are also invited into a three-month extension period to keep building—full-time in-person at the Constellation office, or part-time remote. The stipend continues throughout the extension (prorated for part-time), along with office access and, for full-time in-person extenders, housing.

Yes. Residents receive a $6,000/month stipend. On top of that, Kairos covers housing in Berkeley for the full residency and pays for travel to and from Berkeley — neither comes out of the stipend. Residents invited into the extension period continue receiving the stipend (prorated for part-time), and full-time in-person extenders keep their housing too.

Generator takes place in the Constellation office in Berkeley, California. In-person work is expected for the full duration of the core three months. Kairos covers travel to and from Berkeley and housing in Berkeley for the entire residency, on top of the stipend. Residents invited into the extension period can then either continue full-time in-person or switch to part-time remote. If you have a special circumstance, please apply anyway and let us know.

Yes. We welcome applicants from anywhere in the world, and for most people we'll be able to sponsor J-1 visas for the program's duration (though we can't guarantee obtaining such a visa).

The first two weeks of the program will focus on matching. Kairos and Constellation will organize residents into groups of 1-4, and match each group with a project and a mentor. Residents may work on projects from our list, or on their own ideas.

Generator is designed as a launchpad into full-time generalist roles in AI safety: program managers, fieldbuilders, operators, chiefs of staff, COOs, and founders. During and after the program we actively help residents land these roles, turn their projects into new orgs, or hand them off to existing ones.

Here are a few examples:

Workshops and conferences. Run an impactful domain-specific conference like ControlConf, or one that brings new talent into AI safety, like GCP. Focus on reaching high-leverage, new audiences, or covering emerging subfields of AI safety.

AI comms fellowship. Design and manage a short fellowship for skilled writers and communicators to produce content about AI safety. Draft a curriculum, identify mentors, acquire funding, and prepare a pilot cohort.

Recruiting pipelines. Work with two or three small AI safety orgs and build the systems they need to scale quickly: work tests, candidate sourcing, referral pipelines. Solve recruiting coordination challenges between orgs.

Travel grants program. Design a program to fund visits to AI safety hubs by promising students and professionals. Set admission criteria, build an application flow, line up partner referrals, and run a pilot round.

Shared compute fund. Scope a fund that can rapidly cover the compute needs of independent safety researchers. Model whether a full-on cluster is needed. Acquire compute, deliver a plan, and distribute a pilot round of grants.

Strategic awareness tools. Reduce adversarial pressure during takeoff by scaling AI-powered superforecasting and scenario planning in safety infrastructure. Build support among impactful stakeholders and run a pilot.

Mass human data collection. Build robust systems to collect thousands of hours of human data (RCTs, uplift studies) in just weeks. Work with multiple organizations to scale these systems as part of core research workflows.