AI Security Research

What You'll Do

• Deep-dive into AI usage risks, including threat detection for chat platforms and AI agents.
• Research emerging agent threats and LLM attacks, such as jailbreaks and prompt injection.
• Conduct red teaming and large-scale experiments on AI agents, including quantitative experimentation and advanced result analysis.
• Assess internal and externally-facing AI agents for security posture and risk exposure.
• Evaluate and attack real-time defenses and runtime protections for AI applications.
• Conduct research across cloud, web, and API security to uncover novel threats and attack vectors relevant to AI systems.
• Partner with product and engineering teams to turn research findings into detection logic, guardrails, or product features.
• Stay current on and contribute to industry frameworks (e.g., OWASP LLM Top 10, MITRE ATLAS) — potentially contributing back to the community.
• Publish blog posts, give talks, and represent research at conferences.

Required Qualifications

• Background in AI/ML or security research. Ideal candidates have both, but we're equally open to candidates from data science or security research backgrounds with demonstrated depth in AI systems.
• 5+ years of experience in the AI or security industry.
• Solid understanding of AI agents: how tool use, memory, planning, and orchestration layers interact, and where that architecture introduces risk.
• Working knowledge of LLM and agentic security — prompt injection, jailbreaks, tool-use exploitation, memory/context poisoning, and related attack classes.
• Knowledge of AI vulnerabilities at both the model and infrastructure layers, and familiarity with effective mitigation strategies (e.g., guardrails, sandboxing, input/output filtering, access controls).
• In-depth understanding of LLMs, generative AI, agentic systems, and RAG pipelines — how they're built, where they break, and how failure modes propagate through a system.
• Strong research methodology and judgment: experience designing large-scale experiments, forming hypotheses, and applying quantitative rigor to evaluate results, identify signal from noise, and draw defensible conclusions from complex or ambiguous data.