AI Red Team Trainer
Summary
Trains and calibrates Innodata's global AI red-teaming workforce serving frontier AI labs: designing curriculum and hands-on labs for jailbreaks, prompt injection, and agentic attacks, plus rubrics and calibration sessions. Requires 3+ years hands-on AI red teaming/LLM security and proven teaching experience. Fully remote.
The attacks that still break frontier AI models are multi-turn, obfuscated, persuasion-driven, or embedded in the tools and documents agents interact with. Identifying them at scale requires highly trained red teamers who are calibrated to consistent standards and continually adapt to a rapidly evolving threat landscape.
Innodata’s Agentic Evaluation & Safety team partners with leading frontier AI labs. As an AI Red Team Trainer, you will train and calibrate a global red-teaming workforce, teaching advanced attack techniques, assessing severity and quality, and keeping teams aligned with emerging threats. This is a fully remote position, working alongside senior red teamers, safety researchers, and offensive security engineers. You will report to the Head of AI Cyber Agentic Environments & Testing and help shape a team at the forefront of AI security.
You report to the Head of AI Cyber Agentic Environments & Testing. You will help shape a team at the cutting edge of AI security.
What you'll do
- Build the curriculum. Design structured training from fundamentals to advanced work. Cover multi-turn escalation (Crescendo, context priming, persona drift), cipher and encoding attacks, persuasion taxonomies, many-shot and long-context attacks, and automated methods such as PAIR, TAP, and GCG.
- Teach agentic attacks. Train red teamers on direct and indirect prompt injection, tool and MCP poisoning, memory poisoning, goal hijacking, data exfiltration through agents, and multimodal attacks.
- Run hands-on labs. Build practice environments, attack drills, and capture-the-flag style exercises against real and sandboxed models so people learn by breaking things.
- Calibrate quality and severity. Create rubrics, gold-standard examples, and calibration sessions so every red teamer judges success, severity, and novelty the same way. Measure inter-rater agreement and close the gaps.
- Onboard and certify at scale. Run onboarding for new cohorts across regions and time zones, assess skill levels, and certify people for higher-tier projects.
- Keep the training current. Track new jailbreak research, public red-teaming results, and what is working on live projects, and fold it into training within days.
- Coach and give feedback. Review red teamers' work, give specific, actionable feedback, and identify people ready to move into senior roles.
- Protect the workforce. Build guidance for working safely and sustainably with harmful content, including wellbeing practices and escalation paths.
What you bring
- 3+ years of hands-on AI red teaming, adversarial ML, or LLM security work. You have found real jailbreaks and prompt injections yourself.
- Fluency in the current attack landscape: Crescendo, cipher and encoding attacks, persuasion-based attacks, many-shot jailbreaking, PAIR and TAP, GCG, and indirect prompt injection. You can explain why each works.
- A working understanding of how LLMs and agents behave: system prompts and instruction hierarchy, refusal training, tool calling, and agent loops.
- Proven experience teaching, training, or mentoring technical people, such as running workshops, bootcamps, onboarding programs, or courses.
- The ability to turn tacit expertise into clear playbooks, rubrics, and exercises.
- Experience with quality calibration, such as rubric design, gold sets, and inter-rater agreement.
- Clear, engaging communication in English, live and in writing, with globally distributed teams.
- Mentor junior red teamers, cyber researchers, and RL engineers.
Nice to have
- Experience creating security training content, such as CTF challenges, Hack The Box or TryHackMe labs, or courses for a security training platform.
- Top placements in public AI red-teaming challenges or jailbreak competitions.
- Experience with red-teaming tooling such as Garak, PyRIT, or promptfoo.
- Familiarity with OWASP Top 10 for LLM and Agentic Applications, MITRE ATLAS, and NIST AI RMF.
- Background in trust and safety operations or training large annotation or review workforces.
- Multilingual ability
We’re an equal opportunity employer.
Only relevant applicants will be contacted
Skills
Sign in to read and join the discussion.