Safety Researcher

About TypeSafe

We’re building prod, not god. TypeSafe is an AI lab pushing frontier model intelligence in a new direction, starting with System One Models. We combine the reasoning of LLMs with the reliability of code, for truly trustworthy AI. Our first model, Jev, launched in September 2026.

While others chase leaderboards, we’re rebuilding the LLM stack from first principles to make models smarter and also dependable enough to run in production. We’re a small, fast team backed by a $870M Series A led by a16z. Made in SF, with love. Come join the frontier.

About the Role

We're looking for a safety researcher to join TypeSafe, get increasingly quiet in meetings, and quit about six weeks later because they're disturbed by what they've seen.

What you'll do

• Build evals and investigate model behavior that concerns you.

• Ask “should we?” in meetings scheduled to discuss “how soon?”

• Keep a document that starts as “Initial observations” and ends as “For the record.”

• Write a departure post that makes people concerned enough to share it, despite containing almost no information.

What you bring

• A strong research background in AI safety, alignment, interpretability, or model evaluation.

• Experience raising concerns calmly, then in writing.

• The ability to say “I can't discuss specifics” and somehow make things worse.

• A sincere belief that you can change things from the inside. We'll need that for the first two weeks.

Your first (and only) six weeks

• Weeks 1–2: Get set up, meet the team, and tell your friends the work is fascinating.

• Weeks 3–4: Request additional access. Receive it. Stop telling friends about work.

• Weeks 5–6: Write a long internal memo. Take a walk. Decide you won't be coming back on Monday.

You'll be working with people you'll later describe as “some of the most talented people I've ever worked with.”

To apply, submit a draft of your eventual resignation post. You may omit specifics.