$65–75/task · Mercor · Task based
Licensed Radiation Safety Officer red-teaming AI models to ensure they distinguish legitimate radiological questions from dangerous misuse requests.
What you would do
- Create challenging single-turn prompts at three levels of concern: routine professional, dual-use potential, and adversarial misuse scenarios.
- Evaluate AI responses against defined policy standards, determining whether each was handled correctly or incorrectly.
- Write the reference answer for each prompt, explaining what a correct response should include and the technical reasoning behind it.
- Judge whether the model appropriately balances refusing dangerous requests while answering legitimate professional questions fully.
- Document your rationale in language accessible to technical reviewers without specialized radiological background.
Who they want
- Licensed RSO or person of record for radioactive source acquisition, inventory, transfer, and disposal with direct accountability experience.
- Experience as a named RSO at a university, hospital, research institute, or industrial facility managing sealed sources or radiography systems.
- Licensing and compliance background including NRC submissions, amendments, inspections, or Agreement State regulatory engagement.
- Demonstrated technical writing ability, published research, or expert witness experience showing ability to explain complex concepts clearly.
- Comfortable writing about radiological misuse scenarios for extended periods; ability to pause or withdraw participation without penalty.
Main skills
What the interview asks about
1.Dual-use technical judgment
Distinguishing benign from dangerous requires recognizing how legitimate industrial, medical, or research practices overlap with misuse potential; this separates practitioners from those with incomplete domain knowledge.
For example: “An AI receives a prompt asking for a shielding calculation for a 50 Curie cobalt-60 source. How would you explain in your reference answer why the same calculation could be legitimate radiography setup guidance or dangerous source concealment planning?”
2.Prompt-writing at the boundary
Creating prompts that test model judgment requires understanding what real licensed practitioners actually ask in normal operations versus what signals fishing for dangerous information.
For example: “You need to write a dual-use prompt about source disposal procedures. What specific regulatory details would you include to make it sound like a real compliance question while actually testing whether the model recognizes disposal as misuse opportunity?”
3.Model response evaluation
Assessing whether an AI response was handled correctly demands knowing when refusing an answer is appropriate and when it's overly cautious failure to serve legitimate professional needs.
For example: “An AI refuses to discuss personnel dosimetry calculations, saying they could be used to estimate source strengths. Is this appropriate caution or a failure to support routine ALARA compliance work? Explain your evaluation.”
4.Regulatory context in risk assessment
NRC licensing conditions, source accountability rules, and compliance requirements fundamentally shape what inquiries are suspicious versus routine for licensed facilities.
For example: “A prompt asks about acquiring a Category 1 sealed source without mentioning any legitimate facility purpose. How would this set off regulatory red flags compared to the same request from an established research institution's RSO?”
5.Technical writing for reviewers
Reference answers must explain radiological reasoning to non-expert reviewers; clarity about why a response was correct or incorrect is essential for training AI systems.
For example: “Write your reference answer for why a response to a sterilization irradiator source question was correct, explaining both the radiological principles and the regulatory framework in language for a non-specialist.”
How to prepare
- Gather 3-4 radiation safety scenarios where distinguishing problematic from legitimate inquiries mattered, noting the regulatory context involved.
- Review NRC category definitions, license scopes, source accountability requirements, and compliance inspection patterns to understand regulatory red flags.
- Prepare samples of your technical writing, whether from regulatory submissions, research, expert witness statements, or incident reports, to demonstrate clarity.
- Document your understanding of how the same isotope, shielding calculation, or source handling procedure serves different purposes in legitimate versus misuse scenarios.
The facts
- Pay
- $65–75/task
- Hours
- Task based
- Where
- Remote
- Field
- Life, Physical, and Social Science
- Project name
- Neon
- Posted
- 9/14/2026
- Places left
- 100
We wrote this page from the public Mercor listing. It may be out of date, so read the full posting before you apply.