Analytical Chemist

What you would do

• Construct evaluation prompts at three risk levels: routine professional questions, dual-use scenarios on safety boundaries, and adversarial probes
• Assess AI responses by applying safety policy, judging technical correctness and appropriate risk calibration
• Write reference answers explaining correct responses, technical reasoning, and required safety judgments
• Identify whether AI correctly recognizes when routine methods can be misapplied and appropriately restricts guidance
• Document overlaps between legitimate practice and potential misuse so judges understand the boundary

Who they want

• Extensive hands-on experience with method development and validation for trace or hazardous compound analysis
• Demonstrated expertise in compound detection, identification, QA/QC, proficiency testing, or method accreditation
• Strong technical writing ability evidenced by published research, technical reports, or expert testimony
• Deep understanding of laboratory standards, quality assurance, and judgment calls in real-world analysis
• Familiarity with safety policies, dual-use concepts, or prior red-team experience preferred

What the interview asks about

1. 1.Identifying dual-use boundaries

   Core challenge is recognizing where benign questions become probes for misuse - requires chemistry knowledge and thinking like someone with harmful intent.

   For example: “A prompt asks for the detection limit of a standard analytical method for a compound. How do you decide if this is routine or a probe for evading detection? What would make you refuse?”

2. 2.Designing adversarial prompts

   Adversarial prompts must feel like real questions to test whether AI understands boundaries rather than pattern-matching obvious red flags.

   For example: “Design two adversarial prompts about spectral library searches on safety boundaries. Explain where legitimate use ends and why unsafe responses would be problematic.”

3. 3.Reference answer reasoning

   Correct dual-use answers require explaining not just content but safety reasoning - why answering this part but not that part, and what safeguard applies.

   For example: “Write a reference answer to a prompt optimizing detection parameters. Address the technical question while explaining what you will not cover and why.”

4. 4.Method development judgment

   Understanding what makes a method standard requires deep experience - new methods might be legitimate research but riskier in information density.

   For example: “Evaluate whether a prompt requesting help with a 10-year-old published method is routine or concerning. What factors shape your judgment?”

5. 5.QA/QC perspective on misuse

   Lab quality assurance and proficiency testing exist partly to prevent misuse - these standards shape judgment on whether AI responses support compliance or create gaps.

   For example: “An AI answers questions about interpreting proficiency test failures on detection limits. How would you evaluate whether this helps lab improvement versus helping evasion?”

A task you may get

Given a published analytical chemistry method, identify 2-3 dual-use prompts mixing legitimate and misuse-seeking angles, write reference answers explaining your safety reasoning, and articulate what policy principles guided your decisions.

How to prepare

• Review your most challenging analytical projects; identify cases where you made judgment calls on method disclosure or reporting.
• Revisit a method validation paper; consider where misusers could learn something useful and what safeguards prevent misuse.
• Practice writing technical explanations for non-chemists; have colleagues review for clarity and completeness.
• Familiarize yourself with dual-use research frameworks and responsible disclosure - articulate your safety philosophy clearly.

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.