About HiltLock
HiltLock (previously Trampolyne AI) does AI red teaming for production GenAI: human-led adversarial testing against live AI systems. We find the failures that read as ordinary traffic and never look like a bad request, the classes of attack that scanners, VAPT, and compliance frameworks don't reach. Recent work included pulling 3.49 MB of one company's customer data across tenants on a billion-dollar platform that had passed its pentest and its audit.

We're building a vetted associate pool: experienced security people we bring onto specific engagements, paid per project, on terms set for that project. This is not a full-time role and not a salaried position. It's for people who want engaging, high-signal offensive work on real production AI systems, alongside whatever else they do.

What the work is
You'd be brought in on client engagements to do manual, architecture-aware adversarial testing: reasoning about a specific system's tenancy, tool wiring, and retrieval scope, then hunting attack chains nobody has written a signature for yet. Concretely,



across engagements this looks like:

Designing and running adversarial inputs against live LLM and agent systems to surface prompt injection that survives guardrails, cross-tenant data exposure, tool and function-call abuse, and retrieval and RAG manipulation. Reasoning about authorization boundaries: whether a system can be talked into acting outside its authority. Writing findings that hold up: exploit chains, reproduction steps, severity, and a fix path per finding, not a report the client can't reproduce.

Who we're looking for
Almost nobody in the market has years of AI red teaming behind them, because the discipline is that new. We care about demonstrated ability to break systems and reason about authorization, not a specific title. You're likely a fit if you have 2–3+ years in one or more of:

Offensive security or penetration testing. Security research or vulnerability research. Application secu

📌 AI Red Team Associate (India)
🏢 HiltLock
📍 India