Sr. Ai/llm Threat Researcher, Agentic Systems - AI Detection and Response (hybrid)

CrowdStrike CrowdStrike · Enterprise · Sunnyvale, CA +2

This role focuses on researching and identifying security vulnerabilities in AI agents and LLMs, specifically within the cybersecurity domain. The researcher will analyze agentic systems, LLM architectures, and RAG pipelines to uncover risks like prompt injection and filter bypasses, contributing to defensive strategies and thought leadership in AI security.

What you'd actually do

  1. Vulnerability Research: Conduct deep-dive analysis into LLM architectures and agentic frameworks to identify potential security weaknesses, including risks associated with prompt injection and safety filter bypasses.
  2. Agentic Security Analysis: Evaluate the security boundaries in Agent-to-LLM and Agent-to-Application interactions, focusing on how autonomous loops and multi-step reasoning processes can be secured against manipulation.
  3. Security Assessment: Develop methodologies to test the robustness of RAG (Retrieval-Augmented Generation) pipelines and third-party tool integrations, ensuring resilience against adversarial inputs.
  4. Framework Alignment: Map research findings and defensive strategies to industry standards, such as the MITRE ATLAS framework and the OWASP Top 10 for LLM Applications.
  5. Thought Leadership: Contribute to the security community by publishing whitepapers or presenting research on the safety and security challenges of AI agents at industry conferences.

Skills

Required

  • Comprehensive understanding of transformer architectures, attention mechanisms, and the lifecycle of LLM development.
  • Experience with AI orchestration frameworks and the security implications of autonomous decision-making and long-term memory in AI systems.
  • Comprehensive understanding of LLM Prompts, MCP, A2A and various emerging AI protocols.
  • Knowledge of the evolving LLM risk landscape, specifically regarding insecure output handling, data integrity, and model robustness.
  • Proficiency in Python and experience with AI security evaluation frameworks or custom red-teaming methodologies designed to improve system defenses.

Nice to have

  • A record of academic publications or public research regarding AI/ML security and risk mitigation.
  • Experience in AI red teaming or participating in security evaluation programs.
  • Experience developing defensive layers, such as guardrail systems or monitoring solutions for agentic workflows.

What the JD emphasized

  • security of complex AI Workflows and Agentic Loops
  • multi-step reasoning and tool-calling
  • prompt injection
  • safety filter bypasses
  • Agent-to-LLM and Agent-to-Application interactions
  • autonomous loops
  • multi-step reasoning processes
  • RAG (Retrieval-Augmented Generation) pipelines
  • third-party tool integrations
  • LLM architectures
  • agentic frameworks
  • AI orchestration frameworks
  • security implications of autonomous decision-making
  • long-term memory in AI systems
  • LLM Prompts
  • MCP
  • A2A
  • emerging AI protocols
  • LLM risk landscape
  • insecure output handling
  • data integrity
  • model robustness
  • AI security evaluation frameworks
  • custom red-teaming methodologies

Other signals

  • AI Agents
  • LLM Security
  • Vulnerability Research
  • Adversarial Attacks