AI Safety Red Team Specialist

About OpenTrain

OpenTrain is the #1 platform for people building careers in AI training and data labeling. We help freelancers find specialized projects, build a single verified portfolio of AI training work, and grow long-term freelance careers working on the human side of AI.

About AI training and red teaming

AI training (also called data labeling or human feedback) is how modern models learn from human examples and judgments. Red teaming is a specialized branch of this work: people craft adversarial prompts and multi-turn probes to expose vulnerabilities, bias, and misuse so models can be safer and more robust.

• This role focuses on text-based adversarial testing, structured evaluation, and reproducible attack cases.
• Work directly shapes how conversational AI handles sensitive topics and resists manipulation.

The role

You will work as an AI Safety Red Team Specialist conducting adversarial testing of conversational models and agents. This is remote contract work, part-time at 20+ hours per week, and pays between USD 24 and 35 per hour depending on task complexity and reviewer level.

• Employment type: Contractor, Part-time.
• Hours: 20+ hours/week (flexible scheduling).
• Pay: $24–$35 per hour.
• Work is text-based and focused on evaluation and RLHF-style tasks.

What you'll do

Use playbooks, taxonomies, and benchmarks to run repeatable red-team attacks; annotate and classify failures; and produce clear, reproducible reports and datasets that developers can use to fix issues.

• Design adversarial prompts and multi-turn attack chains to find jailbreaks and prompt-injection vectors.
• Annotate failures, classify vulnerability types, and flag systemic risks.
• Follow structured taxonomies, benchmarks, and playbooks for consistent evaluation.
• Produce reproducible attack cases, datasets, and concise reports.
• Work on sensitive-topic review with provided guidance and safety support.

Requirements

We need testers who bring prior adversarial experience, strong judgment, and clear communication skills. The role expects intermediate-level experience and careful use of structured frameworks.

• Prior red teaming experience in AI adversarial work, cybersecurity, or socio-technical probing.
• Familiarity with jailbreaks, prompt injection, and misuse-case analysis.
• Ability to classify vulnerabilities and explain risks to technical and non-technical stakeholders.
• Comfort using taxonomies, benchmarks, and playbooks to keep testing consistent and reproducible.
• Native-level fluency in English and Thai (required).

Who should apply

Apply if you have hands-on adversarial testing experience, enjoy investigative, structured evaluation work, and want flexible, remote contract work that directly improves model safety. This role suits people who can translate findings into clear, actionable reports.

• Intermediate-level red teamers, cybersecurity professionals transitioning to AI safety, and experienced RLHF annotators.
• People who prefer part-time, remote, project-based work and can commit 20+ hours/week.