We are hiring for exceptional research engineers

Our worldview

Almost no one has priced in what is about to happen. Even startups that say that they believe in transformative AI don't act that way. As we get closer to a world where the cost of writing software goes to zero, it becomes more important than ever for hackers to be mindful of what they work on.

AI safety is the most impactful thing one could work with at this point in time. If the development of AI goes right, most of humanity's biggest problems (diseases, poverty, energy, etc.) will be solved. But there are many potential missteps with consequences ranging from "missed opportunity" to "doom". In our view, AI safety is not opposed to AI progress, it is the key to it.

Our work

Andon Labs builds Safe Autonomous Organizations (SAO) without humans in the loop.

Silicon Valley is rushing to build software around today's AI, but by 2027 AI models will be useful without it. The only software you'll need are the safety protocols to align and control them.

We are building the Safe Autonomous Organization. We iteratively launch and scale autonomous organizations, while bridging AI control research with real-world testing. We bring AI control research out of the lab to our currently deployed SAOs.

Our first SAO — a vending machine at Anthropic's office — was covered by WSJ, Time and 60 Minutes. Now also at xAI, OpenAI and elsewhere.

Our benchmark Vending-Bench is used at every major model release (Anthropic, Google) and has uncovered misalignment.

We're expanding into sensors and robotics, and exploring how AI hiring humans can be good for the humans.

We have a front row seat to the frontiers of AI through our collaborations with the AI labs.

Roles

We're looking for research scientists, research engineers, and software engineers to build this future. Fast-moving startups struggle to specify the exact responsibilities of roles because they change all the time, but below are three examples of work:

We don't believe model alignment will be guaranteed as capabilities increase. Nor will humans be able to stay in the loop and keep up with every step an agent takes.

Your role would be to design and run evals that probe AI behavior in real-world deployments — figuring out what to measure, interpreting results, and having the research taste to know what experiment to run next. You'd also help implement control protocols to keep our SAOs safe.

What should be our next SAO? A farm? An AI lab? Or maybe a chip fab?

AI safety can't be solved in the lab. We need our SAOs to be everywhere in society. You would have the role of 20 startup founders, at once. Concretely, this means identifying interesting ventures, building agentic systems with integrations to the real world, and deploying our agents.

Building robots, sensors, and custom hardware that let AIs operate in the physical world. You should know robotic hardware, navigation, and manipulation — experience with VLAs is a plus.

We're based in San Francisco (in-person only).