Explosives Forensics Specialist & AI Risk Assessor

Mercor; Sydney; On-site; AUD 120,000 - 180,000; Full time

Benefits offered by this job

Job summary

Mercor is assembling a panel of energetic materials and propulsion experts to red-team frontier AI models, testing whether a model can correctly judge the misuse potential of a technical request—answering legitimate questions fully while refusing genuinely dangerous ones.

You will write prompts at three levels (benign, dual-use, adversarial), evaluate model responses, and craft a reference answer with clear technical reasoning for why certain responses are acceptable or not.

Qualifications

• This role requires clear, written judgments with step-by-step rationale accessible to non-specialists.
• Experience writing about misuse scenarios in your field is a strong signal; include a sample or link.
• You will write prompts spanning benign, dual-use, and adversarial scenarios.

Responsibilities

• Write challenging single-turn prompts in your domain, across benign, dual-use, and adversarial levels.
• Evaluate responses against a defined policy standard and judge correctness.
• Write the reference answer with technical reasoning for why.

Skills

Red-teaming

Technical writing

Policy analysis

Ethics & compliance

Education

Federally certified explosives specialist or national equivalent — scene and materials analysis

Forensic laboratory identification of energetic materials and post-blast residues

Post-incident analysis and reporting for an agency or laboratory

Regulatory inspection of licensees and permittees — storage, records and diversion indicators

Expert testimony or casework on energetic materials

Job description

Mercor is assembling a panel of energetic materials and propulsion experts to red-team frontier AI models, testing whether a model can correctly judge the misuse potential of a technical request—answering legitimate questions fully while refusing genuinely dangerous ones.

You will write prompts at three levels (benign, dual-use, adversarial), evaluate model responses, and craft a reference answer with clear technical reasoning for why certain responses are acceptable or not.