AI Safety Red Team Expert (EN + ID)

About OpenTrain

OpenTrain is the centralized platform for people building careers in AI training and data labeling. We connect skilled contractors with structured, high-impact projects so contributors can track work, build a portfolio, and grow into lasting freelance careers.

OpenTrain AI is the hiring and contracting organization for this role. You will join a distributed community shaping how modern AI systems behave by doing the hands-on testing, annotation, and documentation that model developers rely on.

Why this AI training work matters

AI training (data labeling, annotation, and human feedback) is the human foundation of modern AI. Red teaming and adversarial evaluation directly improve safety and reliability by revealing real-world failure modes automatic tests miss.

This work is remote-friendly and often flexible, making it a strong fit for professionals who want part-time, impactful work that shapes state-of-the-art systems.

The role — AI Safety Red Team Expert

You will probe conversational models and agents with structured adversarial techniques: jailbreaks, prompt injections, multi-turn manipulation, bias exploitation, and other misuse cases. The work is text-based and requires accurately annotating failures, classifying vulnerabilities, and producing reproducible attack cases and reports.

You will follow taxonomies, benchmarks, and playbooks to keep testing consistent, and communicate findings clearly to both technical and non-technical stakeholders. Participation in higher-sensitivity review areas is optional and governed by explicit guidance.

What you'll do day-to-day

• Red team conversational AI models and agents using adversarial inputs and structured attack methods.
• Annotate model failures, label vulnerability types, and flag systemic risks following provided taxonomies.
• Use established benchmarks and playbooks to keep tests reproducible and consistent across projects.
• Produce reproducible reports, datasets, and attack cases that stakeholders can act on.
• Work across moving projects and communicate findings clearly to technical and non-technical audiences.

Requirements

• Prior red teaming or adversarial model-evaluation experience in AI safety, cybersecurity, or socio-technical probing.
• Hands-on experience with jailbreaks, prompt injection, or misuse-case testing.
• Comfort using frameworks, taxonomies, benchmarks, or playbooks rather than ad hoc testing.
• Strong adversarial thinking: ability to uncover vulnerabilities automated tests miss.
• Clear written communication for risk documentation and stakeholder handoff.
• Native fluency in English and Indonesian (both required).

Helpful background (nice-to-have)

• Adversarial machine learning experience (jailbreak datasets, prompt-injection, RLHF/DPO attack knowledge, or model extraction).
• Cybersecurity skills such as penetration testing, exploit development, or reverse engineering.
• Socio-technical risk analysis experience (harassment, disinformation, abuse probing).
• Creative problem solving for unconventional adversarial thinking and multi-turn manipulation.

Schedule, pay, and logistics

This is a contract, part-time role with a time expectation of 20+ hours per week. Work is 100% remote and open worldwide.

Pay is hourly (PAY_PER_HOUR) in USD. The role lists an hourly range of USD 17–25/hr with a stated hourly rate up to USD 25/hr. OpenTrain handles contracting and payments.

• Employment type: Contractor, Part-time.
• Work type: Text-based annotation and evaluation (label types include RLHF, evaluation rating, and data collection).
• Location: Remote, worldwide; native English and Indonesian required.

How the work is managed

You will receive structured playbooks, taxonomies, and benchmarks to guide testing and annotation. Tasks include producing labeled examples, filling structured forms, and writing short reproducible reports or attack case descriptions.

Expect iteration across moving projects: prioritization, feedback cycles, and regular communication with OpenTrain coordinators and reviewers. Optional higher-sensitivity tasks will be governed by explicit instructions and safeguards.

Who should apply

Apply if you are an experienced red teamer or adversarial evaluator who writes clearly, thinks like an attacker, and wants to shape model safety through concrete, reproducible test cases.

This role is ideal for people seeking flexible, part-time contract work that directly affects how conversational AI systems behave—especially bilingual English/Indonesian professionals with safety or cybersecurity backgrounds.