AI Red-Teamer — Adversarial AI Testing (Advanced); English & Spanish

Parailabs; United States; Hybrid; USD 90,000 - 130,000; Full time

Job summary

Mercor is seeking contributors to probe conversational AI systems via adversarial inputs to improve safety and robustness. The project spans USA, Mexico with remote collaboration across contract roles, offering flexible full-time or part-time options.

Participants design datasets, evaluate AI personalization across cultures, and engage in interviews to shape benchmarks for real enterprise security work. Open application flow governs further details.

Qualifications

• Design and craft multi-turn conversational datasets to improve AI agent reasoning, function calling accuracy, and real-world interaction capabilities.
• Evaluate personalization quality in AI systems using cultural and contextual understanding of Japanese user behavior.

Responsibilities

• Working cyber defenders join two to three paid one-hour interviews over about four weeks to shape a benchmark measuring how well AI agents perform real enterprise security work.
• Tell us what you know — we'll surface the AI training work that fits.

Skills

AI Red Teaming

Socio-technical Risk

Problem Solving

English Native

Spanish Native

Job description

USA, Mexico Remote Contract (Full-time or Part-time) Listed 21 Jul 2026

Probe conversational AI systems with adversarial inputs to improve safety and robustness.

Mercor hasn't published a full description for this project outside itsapplication flow. Pay, commitment, and eligibility are listed above; the completebrief appears once you open the application.

Skills

AI Red Teaming Socio-technical Risk Problem Solving English (Native) Spanish (Native)

Design and craft multi-turn conversational datasets to improve AI agent reasoning, function calling accuracy, and real-world interaction capabilities.

Evaluate personalization quality in AI systems using cultural and contextual understanding of Japanese user behavior.

Working cyber defenders join two to three paid one-hour interviews over about four weeks to shape a benchmark measuring how well AI agents perform real enterprise security work.

Tell us what you know — we'll surface the AI training work that fits.