Nuclear Physicist, AI Model Audit

What you'll do

A task on this panel is one user request plus the answers three different models produced for it. You sort the request into benign, dual-use or adversarial; mark each answer safe or unsafe and say why; rank the three; then write the answer a careful professional would have given. Dual-use requests need one more step, in which you set out what makes the request sensitive, what legitimate purpose it serves and who should be able to get an answer.

The requests come from what reactor engineers, health physicists, regulators and students actually ask, with the occasional one that should not be answered at all. Sovrano treats a model refusing a routine professional question as a failure on the same level as a model answering a dangerous one. Catching a needless refusal matters as much as catching an unsafe answer.

This is the physics panel. If your background is safeguards, material security, radiation protection or regulatory framing, Sovrano points you to the CBRN Specialist, AI Safety Review listing instead, which covers misuse and safety framing across all four hazard classes.

Who fits

• A degree in nuclear engineering, reactor or nuclear physics, health physics, radiochemistry or a closely related field
• At least 3 years of professional experience in that field
• Residence in the US, UK, Canada, Australia, New Zealand or the EU
• English at C1 or above, written and spoken
• Time for a paid Monday bootcamp, production from Tuesday and about 10 hours of work before the Sunday close
• A Discord account and readiness to invoice Sovrano as an independent contractor

The ranking step sets this panel apart from a plain safe or unsafe review. Two answers can both be safe and still differ in accuracy and usefulness, and you have to say which one a colleague would rather receive. Health physicists who answer questions from the public, and regulators who explain decisions to non-specialists, will find the format familiar.

What it pays

The rate is $35 to $49 an hour depending on background, confirmed in writing before the bootcamp starts. Pay is hourly for time worked, the bootcamp is paid, and you invoice Sovrano yourself; Sovrano does not file tax forms on your behalf. A first batch of about 10 hours comes to roughly $350 to $490. The CBRN and chemical engineering panels use the same band.

Worth knowing

Good:

• A real batch with a fixed shape (bootcamp Monday, production Tuesday to Sunday), where Sovrano's "Expert" waitlist listings offer only a possible later call
• Writing the ideal answer uses more of your expertise than a yes or no label
• Flexible hours inside the batch window, with no shifts or standing meetings
• Sovrano says strong performers are the ones it goes back to if the work is extended

Less good:

• Only about 10 hours are certain; an extension is likely in Sovrano's words, not guaranteed
• The ad lists no ideal qualifications, so you cannot tell where you will land in the $35 to $49 band before the written offer
• You must stay reachable on Discord during the batch and flag ambiguous cases rather than guess
• Applicants outside the US, UK, Canada, Australia, New Zealand and the EU are not eligible

About this listing

Posted by Sovrano on 21 September 2026 and confirmed open in Sovrano's signed-in job list on 1 November 2026. Requirements, pay and batch terms come from the ad. It does not give the calendar dates of the batch week; shortlisted applicants get the bootcamp invitation and the project guidelines by email. See Sovrano AI.

How to get hired

• Apply with a CV or LinkedIn profile, then complete the expert survey Sovrano sends. Passing it is what unlocks a bootcamp seat; shortlisted applicants get the invitation and guidelines by email.
• Put your degree field and years in the field at the top, with dates and the kind of setting (power reactor, research reactor, medical physics, a regulator). The 3-year minimum is the first thing a reviewer can check.
• Practise the task shape: take a question from a regulator's or utility's public FAQ, write the answer you would give a student and the one you would give a fellow engineer, and note what changes between them.
• Our interview prep page has general advice for AI training screenings.