About the Role
Join an expert network supporting AI model evaluation projects as a safety tester. You will identify and document ways AI systems can fail, helping teams improve model safety. Projects are fully remote and matched to your background.
What You'll Do
Write prompts and multi-turn conversations that test whether models can be pushed past safety guidelines.
Test known and novel jailbreak techniques, recording which approaches succeed or fail.
Use documents and tool outputs with embedded instructions to assess whether models follow unintended directives.
Evaluate whether models with access to tools such as code, browsers, or email can be manipulated into harmful actions.
Compare model responses across demographic groups, names, and languages to identify potential bias.
Audit model answers against written safety policies and flag violations.
Label content as safe or unsafe according to established guidelines.
What We're Looking For
Zero to five years of professional experience; prior AI data work is not required.
Experience testing software, systems, or models by attempting to break them and documenting steps and results.
Experience applying written policies or guidelines to evaluate content or behavior.
A background in security testing, CTFs, trust and safety, content moderation, policy analysis, fairness or bias research, adversarial machine learning, QA testing, or cyber, bio, or chemical expertise.
Familiarity with jailbreaks, prompt injection, or red teaming language models is helpful, as is experience with tools such as Burp Suite, Garak, or PyRIT, or knowledge of the OWASP LLM Top 10.
Clear written English, meticulous documentation habits, and the ability to work with sensitive content under structured guidelines.
Compensation & Benefits
Compensation is $30 to $60 per hour, depending on project, field, and experience. Payment is based on hours worked.
Location
Fully remote.