Red-Teaming Quality Assurance Lead (QAL)

As a Red-Teaming Quality Assurance Lead, you will review AI-generated safety evaluations, adversarial prompts, risk analyses, and trainer or QA submissions. You will provide precise written feedback using project guidelines and rubrics, while helping maintain consistent quality across remote contributors.

• Review adversarial prompts, model responses, risk classifications, safety analyses, policy explanations, and vulnerability reports for accuracy, realism, and usefulness.
• Assess adversarial reasoning, policy awareness, safety taxonomy alignment, scenario realism, vulnerability coverage, formatting, instruction-following, and rubric adherence.
• Spot-check trainer and QA submissions, identify recurring quality issues, escalate them, and reinforce quality standards across projects.
• Communicate updates, workflow changes, and review expectations to trainers and QAs through remote collaboration channels.
• Support onboarding, documentation, FAQs, trackers, calibration tasks, and process improvements.
• Help manage contributor activation and follow up when team members are inactive or working inconsistently.

This is a remote, part-time contract role for an expert-level contributor. The role is open to candidates in the United States and requires professional English communication.

• Time: 20+ hours per week.
• Language: Professional fluency in English.
• Experience: AI safety, red-teaming, cybersecurity, trust and safety, content policy, risk analysis, model evaluation, content moderation, or related workflows.
• Knowledge: Adversarial prompting, jailbreak patterns, harmful-content taxonomies, misuse scenarios, policy interpretation, and safety evaluation principles.
• Skills: Careful rubric-based review, excellent attention to detail, strong written communication, and comfort leading or supporting remote teams.
• Tools: Experience with Discord, Google Sheets, Google Docs, trackers, dashboards, or project management systems.
• Additional familiarity: Prompt injection, social engineering, cybersecurity abuse, fraud, self-harm safety, extremist content, misinformation, privacy risk, illicit behavior, bias, and model refusal behavior.