AI Red-Team STEM Evaluator

Stress-test AI systems using STEM knowledge to discover hallucinations, overconfidence, reasoning failures and unsafe technical outputs.

Key responsibilities

• Design adversarial but policy-compliant tests.
• Document reproducible failures.
• Score severity and exploitability within project frameworks.
• Verify fixes through regression testing.

What we look for

• Strong STEM background plus AI evaluation or analytical testing experience.

How success is measured

Unique failures found, reproducibility, severity calibration.

How this works on Jwuma

Candidate profile → skills evidence → domain assessment → calibration task → qualification → project matching → production → peer review → expert QA → performance feedback and progression. Corpshore AI service alignment Applicable across annotation and labeling, RLHF and preference data, model evaluation and red-teaming, multimodal datasets and specialized AI data operations, depending on project scope.

Common questions