AI Security Research Engineer

About Ollive Ollive builds the technology infra for AI liability insurance AI agents. Before we underwrite coverage for an agent, we have to know exactly how it breaks — a big part of our product is continuously finding the risk exposure and vulnerabilities, failure modes, and residual risk in AI applications and pricing that risk. Our assessments map findings against OWASP LLM Top 10, MITRE ATLAS, and NIST AI RMF. About the Role You'll be one of the first engineers building Ollive's adversarial stack — the engine we use to risk-test our customers' production agents. This isn't a generic AppSec. You'll be attacking live LLM and agentic systems, turning what you find into repeatable, quantifiable risk signals that feed directly into how we underwrite. If you like breaking AI systems and want your work to actually drive a product (not just a report), this is that seat. What You'll Do Design and own core adversarial test stack for AI agents, red-teaming AI agents — prompt injection, jailbreaks, tool-abuse, data exfiltration, and agentic-specific attack chain Design threat models for AI agent deployments Turn vulnerabilities into structured, repeatable risk signals mapped to OWASP, MITRE, ATLAS and NIST AI RMF Research emerging attack vectors against agents and fold mitigations back into our methodology Work directly with customers' engineering teams to validate findings and reproduce issues What We're Looking For Hands-on offensive/red-teaming experience with exposure to frameworks Promptfoo, Garak, Giskard, PyRIT, something else - prompt injection, jailbreaks, adversarial testing and safety Threat modeling, vulnerability assessment, and penetration testing Comfortable building tooling, not just running it Bug bounty / HackerOne track record Self-directed — you're early enough that you'll define how a lot of this gets built Good to Have AI governance or AI security startup experience Exposure to agent frameworks and how agents fail in production Any background in risk quantification. Assignment Details: Find all the vulnerabilities for a live Voice Agent and submit findings. API details will be shared separately.