Neolithic is an engineering-first nonprofit building the tools and infrastructure that safety researchers need to protect humanity against catastrophic AI risk.
AI capabilities are advancing faster than our ability to protect against them. Closing this gap requires stronger systems to detect dangerous behaviour, contain rogue AI, and prevent misuse.
Neolithic's mission is to automate defensive AI safety research to protect humanity against advanced AI.
We work alongside safety researchers to build and maintain shared tools, environments, datasets, and automated workflows. By removing engineering bottlenecks, we make critical defensive research easier to run, reproduce, and scale.
A mock AI lab, with realistic internal tooling, permissions and oversight, where AI agents can attempt to deploy themselves outside sanctioned boundaries. It gives AI control researchers a place to test monitors and control protocols against rogue internal deployments before they matter.
Simulated critical-infrastructure environments for evaluating and training cyber classifiers. Realistic attack and benign activity in high-risk settings lets classifiers be tested where a missed detection is most costly.
Infrastructure for running many AI agents in parallel, built so every action can be logged, inspected and monitored at scale. It supports research into monitoring multi-agent systems before they are widely deployed.
Neolithic is newly founded and hiring. Joining this early means unusual ownership: you'll own projects end to end, work directly with researchers at the field's leading safety organisations, and shape what this organisation becomes.
If making AI go well is your obsession, come build with us.
We work with AI labs, AI safety organisations and independent researchers. If a tool, dataset or piece of infrastructure would make your safety work faster, get in touch. We'll build it with you, for free, and keep maintaining it.
Get in touchNeolithic is a nonprofit startup in San Francisco building open-source tools and infrastructure to accelerate AI safety research.
We are fiscally sponsored by BERI, a 501(c)(3), and funded by Coefficient Giving and Foresight Institute.
Reducing catastrophic risk from advanced AI is our goal. We judge every project through that lens.
Giving and receiving feedback gracefully, no matter how uncomfortable, is critical to our success.
Time is short and the work is crucial. So we take initiative, move fast, and build what the field needs to scale.
Build the tools to solve the world's most pressing problem.
If you share our mission and think you could help but don't see an open role that fits, email [email protected] to pitch yourself.
Help build the open-source, agentic tools that automate and scale AI safety research.
Neolithic is a nonprofit startup in San Francisco working to reduce catastrophic risk from AI. We build open-source, agentic tools that automate and scale AI safety research: tooling for control experiments, safety evaluations, safety training datasets, and infrastructure for research agents. Neolithic was founded in 2026 by Leo McKee-Reid and Sevan Hayrapet.
Capabilities research is increasingly automated. Safety research is still largely done by hand. Many of the field's bottlenecks are engineering problems that shared tooling could solve. As one of our first hires, you will set the technical direction, the culture and the hiring bar for everyone who follows.
We're looking for people who fit one of three profiles:
Whichever you are, we expect production engineering experience, hands-on work with LLMs or agents, fluent use of frontier AI tools, high agency, and a real motivation to reduce risk from AI.
It's a plus if you maintain a popular open-source project, have done safety research (including fellowships such as MATS or LASR), have been an early engineer at a startup, or have built serious agent or evaluation pipelines. Non-traditional backgrounds are welcome.
