AI Safety Red Team Specialist, Remote

About OpenTrain

OpenTrain is the #1 platform for building careers in AI training and data labeling. We connect skilled contributors with applied AI-safety work, help people track and grow their portfolios, and make it easy to take on meaningful, paid projects in a single place.

We hire and contract directly for this role. Working with OpenTrain means joining a fast-growing community that helps shape how modern AI systems are tested and improved.

About AI training and red teaming

AI training (data labeling, annotation, and human-feedback work) is the human side of building AI: people craft examples, probe model behavior, and evaluate outputs so models learn to be safer and more reliable.

Red teaming is adversarial testing focused on finding vulnerabilities—jailbreaks, prompt injections, bias exploits and multi-turn manipulation—so developers can fix weaknesses before they reach users.

The role: AI Safety Red Team Specialist

You will adversarially test conversational models and agents using written prompts and multi-turn attacks, classify failure modes, and produce reproducible attack cases, datasets, and reports that help engineers mitigate risk.

This is a part-time contractor role (20+ hours/week) that is fully remote and worldwide. Work is text-based and centers on clear, careful documentation suitable for technical and non-technical audiences.

• Employment type: Contractor, Part-time
• Time requirement: 20+ hours per week
• Pay: $20–$22 USD per hour (PAY_PER_HOUR)
• Data type: Text; label types include RLHF, Evaluation Rating, Red Teaming

What you'll do day-to-day

Follow structured taxonomies, benchmarks, and playbooks to keep testing consistent and reproducible. Produce clear reports and labeled examples that capture how and why attacks succeed.

• Red team conversational AI with adversarial prompts and multi-turn attacks
• Annotate failures and classify vulnerabilities in model outputs
• Flag systemic risks like bias, misinformation, or harmful behaviors
• Document reproducible attack cases and produce datasets and reports
• Work on sensitive topics under explicit guidance and safety rules

Requirements

You must preserve the following skills and attributes from the role description; we will verify language fluency and red-teaming experience during onboarding.

• Prior experience in AI red teaming, adversarial testing, cybersecurity, or socio-technical probing
• Ability to probe jailbreaks, prompt injections, and model misuse cases
• Comfort working with structured frameworks, benchmarks, and playbooks
• Strong written judgment: explain risks clearly to technical and non-technical stakeholders
• Native fluency in English (en) and Assamese (as)

Helpful background (not required)

You’ll stand out if you have hands-on experience in adversarial ML or cybersecurity, or if your background includes socio-technical risk analysis.

• Adversarial ML experience (jailbreak datasets, prompt injection, RLHF/DPO attacks, model extraction)
• Cybersecurity skills such as penetration testing or exploit development
• Socio-technical risk analysis of harassment, disinformation, or abuse
• Creative probing skills from psychology, acting, writing, or similar disciplines

Who should apply

This role is a fit for careful, curious people who enjoy adversarial thinking and structured annotation. It’s suitable for entry-level contributors with prior red-teaming or related experience and for experienced practitioners seeking part-time contractor work.

OpenTrain supports contributors building long-term careers in AI training—this project is a practical way to gain experience working on high-impact AI-safety problems.

How it works: onboarding, pay, and safety

You will complete onboarding that verifies language fluency and domain knowledge and trains you on playbooks and taxonomies we use for consistent labeling.

Pay is hourly at $20–$22 USD and this contract role requires 20+ hours per week. As you contribute, you will produce labeled examples, attack cases, and written reports used to improve model safety.

• Onboarding includes testing and training on playbooks and reporting standards
• Work remotely on text-only tasks and follow strict safety and handling guidelines
• Compensation: paid hourly; exact rate within the listed range depends on assessment