Join AI Safety by doing

The field is hiring. Frontier labs, evals shops, governance institutes, fellowships, and field-building orgs all post open roles every month, and most say the bottleneck is not budget, it is people they can trust to ship.

A representative snapshot of currently-open roles across the field:

• Anthropic · Alignment Researcher, Policy Analyst, Research Engineer, Field Communicator (14 open)
• Apollo Research · Evals Engineer, Research Scientist (8 open)
• Redwood Research · Mech Interp Researcher, Field Builder (6 open)
• FAR.AI · Evals Engineer, Comms Lead (5 open)
• MATS · Program Manager, Field Builder (3 open)
• Constellation · Generator Residency Manager (2 open)

Each role wants the same thing: someone with context in the field, demonstrated skill, and a track record they can verify.

Around 2,000 people complete an AI safety intro course every year. Fewer than 5% land in a structured next step like Generator, MATS, Astra, or Anthropic Fellows.

For nearly everyone else, the same problem shows up in three different shapes:

• No path to practice the work. Intro courses teach concepts. Practicing the actual craft (running evals, writing analysis, briefing policy) needs a team and a real problem.
• No record of contribution. Without shipped artifacts to point to, every job application is a leap of faith for both sides.
• No community of accountability. Self-directed learning burns out without people who notice when you do not show up.

This is the gap Mangrove fills.

Mangrove alumni go on to contribute at AI safety organizations everywhere. Representative placements:

• Coming soon: Apollo Research, Research Engineer (Evals)
• Coming soon: MATS, Field Builder (Programs)
• Coming soon: Anthropic, Policy Analyst
• Coming soon: FAR.AI, Evals Engineer
• Coming soon: Redwood Research, Comms Lead

The field needs contributors in every timezone. Mangrove's overlap-matching means you do not have to move to Berkeley or London to be in the work.

Six features built around the failure modes that kill group projects:

• Candid feedback. Teammates rate the specific role you played. The aggregate per role becomes your skill graph: no five-star averages, no inflation.
• A team that expects you. Four people agreed to do the work together. Three teammates and a kickoff on your calendar do more for motivation than any reading list.
• Zero coordination overhead. Meetings auto-scheduled into the team's overlap window. One repo, one doc, one channel, threaded together.
• A real artifact at the end. Every sprint ships something public at a URL. No demo day theatrics, just the work.
• Sized for normal lives. Two-week project arc. Async-first, with one or two live check-ins. Fits around a job, a degree, a family.
• Mentor drop-ins on tap. Senior researchers opt into one weekly check-in slot. Your team gets a 30-minute risk flag or scope question on Day 5.

AI safety is not one job. It is a network of crafts (research, engineering, comms, policy, mentorship, field-building), each needing different shapes of contribution. The biggest counterfactual impact you can have is not always the role that sounds most prestigious. It is the one where your particular strengths and passions line up with what the field needs.

Through structured peer feedback and prompted introspection over multiple sprints, Mangrove helps you map out where your biggest contributions may be.