AI Red Team Analyst

AI security evaluation

Conduct structured adversarial testing to find ways AI systems can be manipulated or behave unexpectedly. Findings support improvements in model safety, reliability, and security.

• Run red-team exercises that probe weaknesses in AI systems.
• Design adversarial prompts, jailbreak attempts, and unusual scenarios that test the limits of safeguards.
• Assess generated responses for safety concerns, bias, and compliance with applicable policies.
• Document vulnerabilities, exploits, and unexpected behavior in clear reports.
• Recommend practical mitigations with engineering colleagues.
• Track new AI security threats and jailbreak methods, and refine testing criteria and protocols.

Required background

You need a sound understanding of cybersecurity, threat modeling, penetration testing, or vulnerability research, plus practical work with AI or ML systems, LLMs, or prompt engineering. Creative analytical thinking and precise written documentation are essential.

The work requires comfort with independent, asynchronous assignments. Familiarity with open-source AI tools or the OpenClaw ecosystem is helpful; information security, ethical hacking, or AI safety research experience is beneficial but not mandatory.

Contract arrangement

This is a fully remote hourly contract with 10 to 40 hours weekly and flexible scheduling. You work with researchers and engineers internationally while developing further expertise in AI adversarial testing.

Strong performance may lead to broader responsibilities or an extension. The findings feed directly into improvements to AI products used at scale.