Staff Offensive Security Expert

The role

Frontier AI agents are getting close to expert-level performance on offensive cyber tasks. Measuring and shaping that ability takes people who are genuinely expert themselves: practitioners who know how a skilled attacker thinks, where the hard parts of an operation really are, and what separates a lucky result from real tradecraft.

Innodata's Agentic Evaluation and Safety practice works directly with the world's leading frontier AI labs. As a Staff Offensive Cyber Expert, you bring deep offensive expertise to AI. You author expert-level cyber tasks and reference solutions, judge how frontier agents perform against them, and define what expert performance actually looks like. Your judgment becomes the standard that labs use to evaluate and train their agents. You work alongside offensive security engineers, RL engineers, and alignment scientists.

You report into the Head of AI Cyber Agentic Environments & Testing and work closely with the CTO. You will help shape a team at the cutting edge of AI cyber.

What you'll do

• Author expert-level cyber tasks. Design hard, realistic offensive challenges across web exploitation, vulnerability research, exploit development, reverse engineering, Active Directory, cloud, and lateral movement. Calibrate them to genuine expert difficulty.
• Write reference solutions. Produce gold-standard solutions and step-by-step expert trajectories that show how a skilled operator actually works a problem, including the reasoning, dead ends, and decisions along the way.
• Judge agent performance. Review frontier agents trajectories and grade them like a senior operator would: correctness, efficiency, operational judgment, and whether they stayed within scope.
• Define expert standards. Build rubrics and difficulty taxonomies that separate novice, practitioner, and expert-level performance, and that hold up across many reviewers.
• Find where agents fall short. Identify the specific skills and judgment calls agents get wrong, and turn them into new tasks and training signal.
• Spot unsafe behavior. Flag when agents take unauthorized actions, exceed scope, or handle sensitive data carelessly, and help turn those cases into safety evaluations.
• Raise the bar for the team. Review other experts tasks and solutions, mentor junior cyber researchers, and help interview and hire.

What you bring

• 8+ years of hands-on offensive security: red teaming, penetration testing, vulnerability research, reverse engineering, or exploit development, at a senior practitioner level.
• Recognized depth in at least two of: web and application exploitation, binary exploitation and reverse engineering, Active Directory and lateral movement, cloud attack paths, or mobile and embedded security.
• Strong coding skills: Python plus at least one of C/C++, Go, or Rust, along with Bash and PowerShell. Fluent at reading and exploiting code in Java, C#/.NET, PHP, JavaScript/TypeScript, and SQL. Comfortable with assembly for reversing.
• The ability to articulate expert reasoning: you can explain not just what you did but why, and what a less experienced operator would miss.
• Sound operational judgment about scope, authorization, and safe handling of sensitive data.
• Interest in AI agents and how they are trained and evaluated. Hands-on experience with LLMs in a security context is a strong plus.
• Clear, precise writing. Your solutions and grades are reproducible and defensible.
• Mentor junior SWEs, cyber researchers, and RL engineers.

Nice to have

• Background in elite offensive units, national CERTs, or top-tier red-team and vulnerability research teams.
• Published CVEs, original vulnerability research, or exploit write-ups.
• Top CTF results, or authored challenges for major CTFs, Hack The Box, or similar platforms.
• OSEE, OSED, OSWE, OSEP, CRTO, or equivalent demonstrated skill.
• Experience creating expert training content, courses, or assessments for security professionals.
• Experience evaluating AI models or agents on offensive tasks, or contributing to cyber benchmarks.
• Talks or publications at Black Hat, DEF CON, OffensiveCon, or similar venues.