AI Safety Red Team Specialist (English & Bengali)

About OpenTrain

OpenTrain is the centralized platform where people build careers in AI training and data labeling. We connect experienced contributors with specialized projects, help you consolidate proof of work into a single profile, and support long-term freelance careers in an industry that’s growing fast.

Working with OpenTrain means you join the human side of AI development: the skilled people who teach, evaluate, and harden models so they behave more safely and usefully.

About AI Training and Red Teaming

AI training (also called data labeling or human feedback work) is how modern models learn from human examples and corrections. Red teaming is a specialized subset focused on adversarial testing: finding ways models fail, exposing jailbreaks and prompt-injection attacks, and producing reproducible cases that engineers can fix.

This role is text-focused and may require reviewing sensitive outputs (bias, misinformation, harmful behavior) under clear guidance and safety rules.

The Role — What You’ll Do

You will perform adversarial evaluations of conversational AI: craft and run prompts, pursue multi-turn manipulation, document failures, and produce clear, reproducible attack cases that feed model improvements.

• Probe conversational AI systems with adversarial inputs and multi-turn manipulation.
• Annotate model failures and classify vulnerabilities using provided taxonomies and playbooks.
• Document reproducible attack cases and produce clear written findings for technical and non-technical stakeholders.
• Follow benchmarks, structured frameworks, and project playbooks to keep evaluations consistent.

Requirements

This is an expert-level red teaming assignment. You must demonstrate prior red teaming, adversarial AI testing, cybersecurity, or socio-technical probing experience, and the ability to explain risks clearly in writing.

• Prior red teaming or adversarial AI testing experience (required).
• Familiarity with jailbreaks, prompt injections, and model vulnerability probing.
• Ability to classify failures and explain risks clearly in written English.
• Fluent reading and writing in English and Bengali (required).
• Comfort using structured frameworks, benchmarks, and playbooks and adapting to changing project instructions.

Helpful Background

The following experiences are useful but not strictly required; they will make you more effective on complex red-team tasks and may be used to prioritize assignments.

• Adversarial ML experience: jailbreak datasets, prompt-injection work, RLHF/DPO attack knowledge, or model extraction familiarity.
• Cybersecurity experience: penetration testing, exploit development, or reverse engineering.
• Experience analyzing harassment, disinformation, abuse, or conversational AI testing at scale.

Compensation, Schedule, and Logistics

This is a part-time contractor role requiring 20+ hours per week. Pay is hourly at $20–22 USD per hour, paid per hour worked.

Work is text-based and can be done remotely from anywhere (worldwide). You will follow project-specific instructions and use provided taxonomies and playbooks to keep results consistent.

• Payment: Pay-per-hour, USD $20–22/hr.
• Time requirement: 20+ hours per week (part-time contractor).
• Data type: Text; label types include RLHF, evaluation rating, and red teaming.

Who Should Apply

Apply if you are an experienced red teamer or adversarial tester who writes clearly, works reliably with structured frameworks, and is fluent in English and Bengali. This role fits people who want impactful, part-time AI-safety work and who can adapt as evaluation priorities evolve.

How It Works

Create or update your OpenTrain profile to highlight red teaming and adversarial AI experience and your language skills. Use the platform to apply, track assignments, and build a portfolio that demonstrates your safety testing work.

Successful contributors produce reproducible attack cases and well-documented findings that training teams can action. OpenTrain helps you discover more opportunities and grow a durable freelance career in AI training.