Solumn AI is a niche Safety & Security partner for frontier AI labs. We build the datasets, evaluations, and environments that help frontier AI labs make their models safer, more secure, and better aligned.
We work in the settings where model behaviour is hardest to get right — agentic systems, cybersecurity, and tool use — building training data that targets specific failure modes, sandboxed environments where a model's behaviour can be measured or trained, and independent evaluations a safety team can rely on.
