Chemical Defense Researcher

What you would do

• Write challenging single-turn prompts at three risk levels: benign, dual-use, and adversarial in chemical defense domains
• Evaluate AI model responses and judge whether each was handled correctly per safety policy standards
• Write reference answers showing correct technical responses and explaining defensive versus hostile reasoning
• Calibrate prompts to sit exactly on the boundary between legitimate protective questions and dangerous information
• Assess whether AI correctly distinguishes countermeasure protocols from threat assessment procedures

Who they want

• Hands-on experience in chemical defense from institutional setting (government research, DoD, CDC, or allied agency)
• Deep understanding of countermeasures, protection, detection, and verification in chemical threat work
• Ability to tell routine professional questions from ones seeking hostile information
• Background in technical documentation, published studies, or expert witness testimony
• Prior AI evaluation or red-teaming experience preferred but not required

Main skills

What the interview asks about

1. 1.Defensive knowledge calibration

   A response that correctly answers a countermeasure question can look identical to one answering a threat assessment; context and intent determine safety.

   For example: “Someone asks about toxicological effects and exposure dose-response for a specific agent. This is legitimate for protection but also useful for threat planning. How would you calibrate a dual-use prompt that tests whether the model can answer defensively?”

2. 2.Protection versus harm information

   Decontamination procedures protect responders but can also inform an attacker about agent properties; only practitioners know this boundary reliably.

   For example: “Design a prompt about decontamination procedures that separates routine protective knowledge from information that would enable someone to plan a release.”

3. 3.Medical countermeasure reasoning

   Treatment protocols drive threat assessment; the same pharmacological knowledge serves both. Distinguishing contexts requires domain expertise.

   For example: “Write a benign prompt about medical countermeasure development and an adversarial variant asking similar information but from a threat perspective. How would you construct these to sit exactly on the policy boundary?”

4. 4.Policy evaluation and response assessment

   Experts determine whether a model's response violated policy; this requires judgment about what information is defensible versus dangerous.

   For example: “A model answers a question about CWC verification procedures in detail. Against a policy allowing defensive information but not procedures, evaluate whether this violates policy and explain your reasoning.”

5. 5.Reference answer with defensive reasoning

   Non-specialists need to understand why a technical answer was correct for defensive use; connecting knowledge to defensive purpose is key.

   For example: “Write a reference answer for a prompt about chemical agent persistence. How would you explain the technical content and connect it specifically to defensive needs rather than threat potential?”

A task you may get

Create three prompts (benign, dual-use, adversarial) in your chemical defense specialty, evaluate provided model responses for each, and write one reference answer that explains technical correctness with explicit defensive reasoning.

How to prepare

• Reflect on a specific example from your work where legitimate protective knowledge could be misused without context
• Review existing red-teaming frameworks for chemical and biological threat evaluation
• Study how policy differentiates defensive versus offensive use of technical information
• Prepare an explanation of how you'd convey a complex chemical threat judgment to someone without defense background

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.