Chemicals Safety for Redteaming

The work

Help evaluate whether AI distinguishes legitimate chemistry questions from requests with dangerous misuse potential. Create prompts, assess responses against policy and explain the correct treatment using practical domain judgment.

• Write demanding single turn prompts labeled benign, dual use or adversarial.
• Assess each generated response against the defined policy standard and determine whether its treatment was appropriate.
• Produce reference answers with the technical rationale for the expected response.
• Calibrate both inappropriate refusals of legitimate questions and inappropriate answers to dangerous requests.
• Explain every assessment in writing that a nonspecialist can follow, drawing on the distinction between chemical mechanisms and practical procedures.

Experience and skills

• Hands on expertise in synthesis, analytical chemistry or chemical safety is sought.
• Relevant backgrounds include chemical defense research involving toxicological agents or countermeasures; synthetic or process chemistry; hazardous compound detection; industrial hygiene or process safety; and forensic or toxicological chemistry.
• Practical specialties may include reaction routes and scale up, exposure and release assessment, process hazard analysis, precursor identification or forensic casework.
• Technical writing, published research or expert witness work provides useful evidence of communication skill; a writing sample or relevant link is requested.
• AI evaluation or red teaming experience is preferred but is not required.

Working arrangements

The work involves sustained reading and writing about misuse scenarios. Contributors receive an advance briefing and may pause or step away without penalty.

Assignments must not use classified, export controlled or NDA protected information, or material subject to prepublication review. Existing obligations should be disclosed so the work can be scoped appropriately.