Jason Koeller

I lead the Applied White-Box Methods team at FAR.AI. We focus on developing, evaluating, and demonstrating methods which leverage model internals to improve the safety of AI systems. We work on diverse AI safety applications of white-box methods, from white-box control to eval awareness to shaping training dynamics.

Black-box methods – such as chain-of-thought monitoring – work well for now. But we are quickly entering a world where black-box interventions and monitoring are insufficient. Interpretability research is still often early-stage, curiosity-driven work without realistic evaluations on applications that matter. My team is bridging the gap between exploration and deployment by stress-testing white-box methods on real-world tasks and using this feedback loop to enable the deployment of better white-box methods at frontier scale.

I am hiring! If you’d like to work together, please apply for a research scientist (here) or research engineer (here or here) position and mention that you’re interested in Applied White-Box Methods.

Formerly an Anthropic AI Safety Fellow (January–May 2026), where I worked with Fabien Roger on building production-ready sandbagging probes.

Before that, I was co-founder and CTO at Chemix, a startup which used ML to accelerate the development of next-generation batteries for electric vehicles. I led all software and ML at Chemix, including both research and engineering. Chemix raised more than $30M in VC financing and employed more than 20 people. Chemix is now Teserac.ai.

I have also worked as a data scientist and a battery scientist, and I earned a PhD in theoretical physics (quantum gravity) from Berkeley.

I’ve dedicated my professional career to research on impactful societal problems, from technical mitigations for climate change to AI safety and alignment.