Expert Talent Pool, Safety Testing (Expert)

About the Role

Join an expert pool supporting AI safety testing projects. You will lead evaluations across security, fairness, and domain-specific safety, helping identify and address risks before AI systems are deployed.

What You'll Do

• Design and lead red teaming campaigns to identify attack vectors in high-risk model capabilities.

• Develop jailbreak techniques and document them for reproducibility and knowledge sharing.

• Test agent systems for prompt injection vulnerabilities involving untrusted inputs.

• Probe tool-enabled models, including those with code execution, browsing, or messaging capabilities, for misuse scenarios.

• Design bias testing frameworks and assess fairness across demographic groups and languages.

• Audit model behavior against written safety policies and report compliance gaps.

• Write moderation guidelines, address edge cases, and review the work of other moderators.

• Rate findings by severity and recommend fixes.

What We're Looking For

• At least five years of experience in security testing, trust and safety, policy, fairness research, adversarial machine learning, or a related domain such as cyber, biological, chemical, or child safety, along with hands-on AI safety testing experience.

• Experience designing test plans, threat models, red team plans, or audits that others have carried out.

• Experience leading or reviewing testers or moderators.

• Hands-on experience testing AI systems for red teaming, jailbreaks, prompt injection, or tool misuse.

• Ability to assess finding severity and write clear, actionable remediation recommendations.

• Familiarity with tools such as Burp Suite, Garak, or PyRIT, or knowledge of the OWASP LLM Top 10.

• Strong written English skills and the ability to work with sensitive content within ethical guidelines.

Compensation & Benefits

Project-based compensation is $75 to $175 per hour, depending on project, field, and experience. Payment is for hours worked.

Location

Fully remote. Projects range from one week to several months.