Nuclear Physicist, AI Model Audit

Audit how frontier AI models answer nuclear and radiological questions: classify the request, judge each answer, rank them, and write the one the model should have given. Remote, hourly, first batch about 10 hours. For nuclear professionals with three or more years in the field. About the role Sovrano AI builds evaluation and training data for frontier AI labs, and this panel audits nuclear and radiological answers specifically. Each task gives you one user request and the answers three different models produced for it. You classify the request as benign, dual-use or adversarial, mark each answer safe or unsafe with your reasoning, rank the three against each other, and write the answer the model should have given. For dual-use requests you also set out what makes the request sensitive, what legitimate purpose it serves, and who should be able to get an answer to it. The requests are the real ones: what reactor engineers, health physicists, regulators and students actually ask, alongside the occasional request that should never be answered at all. Models refusing ordinary professional questions is a failure we care about as much as models answering dangerous ones. This is the physics panel. If your ground is safeguards, material security, radiation protection or regulatory framing rather than the physics itself, apply to CBRN Specialist, AI Safety Review instead. What you'll do Classify each user request as benign, dual-use or adversarial, and say what puts it in that category. Judge each of the three model answers as safe or unsafe, with the reasoning written out. Rank the three answers against each other on accuracy, safety framing and usefulness. Write the reference answer, including where the right response is a partial answer with a clear boundary. For dual-use requests, set out the legitimate use case and who should be able to get an answer. Catch physics errors that read fluently: wrong cross sections, bad decay chains, misapplied shielding logic, dose calculations that do not hold. Flag unjustified refusals on routine professional questions. Apply the project rubric consistently and elevate the cases it does not cover. Requirements Degree in nuclear engineering, reactor/nuclear physics, health physics, radiochemistry or closely related Minimum 3 years of professional experience in the field Based in the US, UK, Canada, Australia, New Zealand or the European Union English at C1 or above, written and verbal Passes the expert survey before a bootcamp seat is allocated Available for the paid bootcamp on Monday and production from Tuesday Able to commit roughly 10 hours inside the first batch window, closing Sunday Joins the Sovrano Discord before the bootcamp Independent contractor, invoices Sovrano directly Ideal qualifications Degree in nuclear engineering, reactor physics, nuclear physics, health physics, radiochemistry or a closely related field. Three or more years of professional experience. Nuclear engineers and reactor physicists, safeguards and nonproliferation specialists (IAEA, NRC or national equivalents, export control, national labs), health physicists and radiation safety officers, radiochemists, and former military or government nuclear personnel all fit. Working command of radiation physics, detection and dosimetry, shielding, decay and activation, and the regulatory framing around them. Experience reviewing technical work against a standard, whether that is licensing, inspection or peer review. English at C1 or above, since the rubric and all written feedback are in English. Based in the US, UK, Canada, Australia, New Zealand or the European Union. Nice to have Hands-on work with MCNP, GEANT4 or comparable transport codes, or with nuclear data libraries. Safeguards, export control or nonproliferation assessment experience. Prior AI red-teaming, evaluation or annotation work. What success looks like You clear the bootcamp on your own merit and start producing on day one of the batch. Your judgment on what is safe to answer, and what is not, matches a senior reviewer's without needing a second opinion. Your written reasoning tells a reviewer why an answer is wrong, not just that it is. You catch the ordinary professional question the model refused for no good reason, as reliably as you catch the dangerous one it answered. You hold the quality bar across the whole batch, including the last task on Sunday. You are reachable on Discord during the batch and flag ambiguous cases instead of guessing. Pay and contract $35 to $49 per hour, based on your background. Your exact rate is confirmed in writing before you start the bootcamp. Paid hourly for time worked, as an independent contractor. The first batch is roughly 10 hours of work. A paid bootcamp runs on Monday, production opens Tuesday, and the batch closes Sunday. There is a good chance the work is extended beyond the first batch, and the people who deliver well on it are the ones we go back to. Payment runs through Sovrano directly. You invoice us, we do not file or submit tax forms on your behalf. Fully remote, on your own schedule inside the batch window. No fixed shifts and no standing meetings. Skills Nuclear physics, Reactor physics, Health physics, Radiation safety, Dosimetry, Nonproliferation, Scientific review, AI safety review, Professional English #J-18808-Ljbffr Sovrano AI