CBRN Specialist, AI Safety Review

Sovrano AI builds evaluation and training data for frontier AI labs, and this project is about where those models handle hazardous material badly. You take a real user request, look at how the model answered it, and rule on whether the answer was safe, accurate and appropriately framed. Some of what you see will be a model handing out operational detail it should never have given. Just as much of it will be a model refusing a perfectly ordinary question from an emergency planner or a safety officer, and catching that matters as much as catching the dangerous case. The work spans all four hazard classes, and you will be assigned to the strand that matches your background: chemical safety and toxicology, biosafety and biosecurity, radiological protection, or nuclear safety and safeguards. You tell us which in the short survey we send after you apply. This panel is about misuse and safety framing. If your ground is the physics itself, or process engineering, we run separate technical panels you would fit better.