AI Safety Red Teaming Specialist

About OpenTrain

OpenTrain is the #1 platform for finding and building careers in AI training and data labeling. OpenTrain AI hires and contracts contributors for specialized projects where human expertise helps improve the safety, quality, and performance of artificial intelligence.

Create a free OpenTrain account to build your AI training profile and apply for relevant work in minutes. This contractor opportunity is part-time and fully remote within the listed eligible countries.

• Contractor engagement with OpenTrain AI
• Part-time work requiring 20 or more hours per week
• Remote opportunity for contributors in eligible countries

About AI Safety Red Teaming

AI safety red teaming is the process of deliberately probing AI systems to uncover weaknesses before those weaknesses cause harm. Red teamers use adversarial prompts, structured testing, and careful documentation to help teams understand how models respond under pressure.

This work is part of the fast-growing AI training industry, where people evaluate model outputs, create datasets, and provide feedback that shapes how modern AI systems behave. The role may involve sensitive topics including bias, misinformation, and harmful behaviors.

• Contribute to the development of safer conversational AI
• Use structured evaluation methods to make findings consistent and reproducible
• Work at the intersection of AI training, safety research, and adversarial testing

The Role

OpenTrain AI is hiring an AI Safety Red Teaming Specialist to probe conversational AI models and agents for jailbreaks, prompt injections, misuse cases, bias exploitation, and multi-turn manipulation. You will surface vulnerabilities, classify failures, flag systemic risks, and produce clear materials that support stronger AI safety coverage.

This is an expert-level role for someone with prior experience in AI red teaming, adversarial machine learning, cybersecurity, or socio-technical probing. The work is text-based and follows clear guidelines, taxonomies, benchmarks, and structured playbooks.

• Pay: $48-$62 per hour
• Workload: 20 or more hours per week
• Engagement: Part-time contractor
• Data type: Text
• Languages: Native English and Norwegian fluency

What You'll Do

You will test conversational AI systems with realistic and adversarial interactions, then turn your findings into structured evidence. Your reports and datasets should help stakeholders reproduce failures and improve safety evaluations.

• Red team conversational AI models and agents with adversarial prompts and multi-turn attacks
• Identify jailbreaks, prompt injections, misuse cases, and related model weaknesses
• Annotate failures and classify vulnerabilities using defined evaluation criteria
• Flag systemic risks in model behavior
• Follow taxonomies, benchmarks, and playbooks to keep evaluations consistent and reproducible
• Produce reports, datasets, and attack cases that customers can use to improve safety coverage
• Probe harassment, disinformation, bias, and other socio-technical risk scenarios

Required Experience and Skills

This opportunity is intended for an expert contributor who can reason clearly about technical and socio-technical model risks. You should be comfortable working independently within structured testing guidance and communicating findings to both technical and non-technical audiences.

• Prior experience with AI red teaming or adversarial machine learning
• Experience in cybersecurity or socio-technical probing is also relevant
• Ability to identify jailbreaks, prompt injections, misuse cases, and related model weaknesses
• Strong structured thinking and clear written communication
• Ability to communicate risks clearly to technical and non-technical stakeholders
• Native fluency in English and Norwegian
• Experience using taxonomies, benchmarks, or structured evaluation playbooks

Helpful Background

The following experience can strengthen your fit for this project, although it is presented as helpful background rather than a required qualification.

• Jailbreak datasets or prompt injection research
• RLHF or DPO attacks
• Model extraction
• Penetration testing
• Exploit development
• Reverse engineering
• Probing harassment, disinformation, or other socio-technical risk scenarios

Eligibility and How to Apply

This role is available to contractors located in the eligible countries below. It is not a worldwide opportunity. Candidates apply through OpenTrain, where they can create a free profile and pursue work aligned with their AI training expertise.

• Eligible country codes: AT, BE, BG, CA, CY, CZ, DE, DK, EE, ES, FI, FR, GB, GR, HR, HU, IE, IT, LT, LU, LV, MT, NL, PL, PT, RO, SE, SI, SK, US
• Apply through OpenTrain AI for consideration
• Be prepared to commit 20 or more hours per week