AI Agent Quality Engineer

Join the Future of Security at Netskope Netskope (NASDAQ: NTSK) is a leader in modern security and networking for the cloud and AI era. We secure and accelerate cloud, data, and AI in real time, everywhere. Thousands of customers, including more than 30 of the Fortune 100, trust the Netskope One platform, its Zero Trust Engine, and the powerful NewEdge network to gain full visibility and control without performance trade-offs. At Netskope, our technology is driven by our greatest strength: our people. We believe that belonging powers innovation, and success is both personal and organizational. We embrace differences in gender, ethnicity, beliefs, ability, and identity, creating an environment where every voice is heard and respected. We empower our employees to bring their authentic selves to work, grow their careers through continuous education and mentorship, and lead with transparency and curiosity. Join a team where you belong, where you are encouraged to be an entrepreneur, and where together, we continue to redefine the landscape of security. Visit Careers at Netskope to learn more. Follow us on LinkedIn and Instagram . As a Senior AI Quality & Red Team Engineer at Netskope, you will lead the charge in testing, stress-testing, and breaking our AI agents before they ever reach production. From automating multi-turn prompt injections to tracking fleet-wide drift in CI/CD, you will own the automated harness that ensures our AI systems are secure, resilient, and compliant. If you love the idea of being the person who proves an agent isn't ready yet, welcome home. Skills and competencies: Build and grow the automated evaluation suite every agent runs against before it's approved for production, designed to run unattended and scale across a growing agent fleet, not something that needs a person babysitting each run. Design adversarial test scenarios — prompt injection attempts, sycophancy checks where an agent has to correctly push back on a false premise, multi-attempt attacks rather than single-shot ones — and automate them so they run on every relevant change, not just before a big release. Own the "break it on purpose" pass for every new agent: attempt to extract data it shouldn't expose, get it to act outside its registered tool boundaries, or get it to treat a synthetic test probe as real. As the fleet grows, build this into a repeatable, scriptable process rather than a manual exercise redone from scratch each time. Partner with the Data Steward on data sensitivity classification for the systems agents touch, so your test scenarios reflect what's actually at stake, not a generic checklist. Decide, for each agent capability, what "pass" actually means, and build that judgment into automated thresholds wherever possible so evaluation keeps up as the number of agents climbs into the hundreds. Maintain the guardrail and negative-test catalog (fail-closed vs. fail-graceful behavior) across the platform, and add new cases as new failure modes get discovered in the wild. Produce clear, audit-ready evidence for every agent's evaluation results, generated automatically as part of the pipeline rather than assembled by hand for each review. Track drift over time across the whole fleet, not agent by agent, so a slow-moving problem in one corner doesn't go unnoticed just because no one's looking at that specific agent that week. Must-Have: At least 4 years in software quality, security testing, or a related discipline, with 1–2 years specifically evaluating or red-teaming LLM-based systems — not just running unit tests against traditional code. Strong Python skills, since the evaluation harness, adversarial test scripts, and automated pipelines will mostly be built in it. Comfortable writing production-quality code, not just glue scripts. Working knowledge of REST APIs and webhook/event-driven patterns, enough to build test harnesses that call an agent's tools directly and validate its inputs and outputs, not just its final chat re…