Handshake
AI Red Teamer (Remote)
Remote (USA) · Remote · Full-time · Posted Oct 5, 2026
About the job
Stress-tests large language models with adversarial prompts, evaluates harmful or incorrect responses, and documents findings to help research labs improve model safety and robustness.
What you'll do
- Create adversarial prompts and scenarios to probe model safeguards across risk categories and modalities.
- Score model responses using harm taxonomies and severity rubrics, and document experiments and findings.
- Collaborate with researchers and engineers to improve defenses and refine evaluation methods.
What they ask for
- Strong hands-on experience using multiple LLMs.
- Strong written communication and ethical judgment.
- Creative, adversarial problem-solving skills.
- Self-directed and collaborative, with comfort working in feedback-heavy environments.
Nice to have
- Familiarity with Python or other scripting languages, LLM APIs, evaluation tools, or rubric-based annotation.
- Relevant experience in trust and safety, moderation, QA, security research, or a high-risk subject area.
Pay
$32 – $95 per hour (hourly pay)
Tools
ChatGPT, Claude, Gemini, Python
AI focus
Adversarial testing of LLM safety and robustness, including jailbreaks, prompt injection, harmful outputs, bias, hallucinations, and other model vulnerabilities.
Model-written summary · the original posting is authoritative