Skip to content
Red Cell

Careers · Open role

Offensive Security Consultant — AI Red Teaming

Test the applications, agents and integrations around deployed AI systems — and the web, API and cloud surfaces they sit on — under written authorization, and turn what you find into evidence a client can fix.

The work

Red Cell assesses what AI systems can be made to do with the access they have been given: assistants that read customer text, agents that call internal tools, retrieval systems that reach document stores. Most of that risk sits in the application and permissions around the model, not in the model itself — and most of it has never been tested as an access path.

You would scope and run those assessments alongside conventional web application, API and cloud testing, because an AI finding is rarely only an AI finding. Every engagement runs under a signed scope, rules of engagement and written authorization to test.

What you would do

  • Help scope engagements: map the deployed architecture, the trust boundaries and what each component can reach, before any testing starts.
  • Test for direct and indirect prompt injection, unsafe tool use, data exposure across retrieval boundaries, and the application controls meant to contain them.
  • Test the surrounding web applications, APIs and cloud identity paths the AI system depends on.
  • Write findings that hold up: reproducible evaluation cases, evidence captured without retaining production customer data, and remediation guidance aimed at the control rather than the single request.
  • Walk client engineers through the results, and retest the agreed findings once they have been fixed.

What we are looking for

  • Hands-on experience in penetration testing or application security assessment.
  • A working understanding of how LLM applications are built: prompts, retrieval, tool and function calling, agent loops, and where each one trusts its input.
  • Comfort with web and API testing, and enough cloud identity knowledge to follow a path.
  • Clear, precise writing. The report is the product the client keeps.
  • An unconditional commitment to working inside written authorization and an agreed scope.

Useful, not required

  • Published research, conference talks, or public tooling in offensive security or AI safety.
  • Experience with adversary simulation or purple team exercises.
  • Background building software, particularly LLM-backed features.

How we work

Nothing is tested until the scope, rules of engagement and authorization are signed. Critical findings are raised when they are found, through the agreed escalation contact, not saved for the report. Engagement material is handled under the terms agreed for that engagement.

All open roles

Start with the system you are least comfortable about.