About the job
As AI systems—especially agentic and autonomous AI—become deeply embedded in our products and internal platforms, the security model must evolve. Traditional application security alone is no longer sufficient. We are looking for an AI Red Team Engineer to help us proactively identify, understand, and mitigate AI-native and agent-specific security risks before they reach production. In this role, you will build and execute adversarial red-teaming exercises against AI models and AI agents, focusing on how they can be manipulated into unsafe, unintended, or harmful behavior. You will work closely with AI platform teams, product engineers, and security partners to stress-test agent logic, tool usage, memory, and autonomy—and translate findings into concrete guardrails and defenses. This role is ideal for someone who enjoys thinking like an attacker, understands modern AI systems, and wants to work at the intersection of security, AI, and real-world impact.
Responsibilities
Design and execute AI red-teaming exercises against LLMs and AI agents, including:
1. prompt injection (direct & indirect)
2. jailbreaking and policy bypass
3. model and tool poisoning
4. memory and context poisoning
5. behavioral drift and unsafe autonomy
6. tool misuse and emergent privilege escalation
Analyze agent workflows, logic, and tool graphs to identify systemic security weaknesses beyond prompt-level attacks.
Develop reusable adversarial test cases, attack libraries, and red-team playbooks for AI systems.
Collaborate with AI platform and product teams to translate red-team findings into actionable mitigations, guardrails, and design changes.
Qualifications
Minimum
1. 3+ years of experience in security engineering, offensive security, red teaming, or AI security.
2. Hands-on experience red-teaming AI models or AI agents, including testing for prompt injection, jailbreaks, unsafe behavior, Excessive agency, Model DoS.
3. Strong understanding of security fundamentals (threat modeling, secure design, least privilege, defense in depth).
4. Ability to clearly document findings and communicate risk to both technical and non-technical stakeholders
5. Proficiency in at least one programming language (e.g., Python, Go, Java, or similar)
Preferred
1. Familiarity with AI security tools and frameworks (e.g., PyRIT, AgentDojo, Promptfoo, custom harnesses).
2. Good understanding of GenAI and LLM architectures, including: embeddings, RAG, or agent frameworks.
3. Hands-on experience executing AI Red Teaming exercises, including prompt injection/jailbreaking, unsafe behavior/behavioral drift, model/tool poisoning.
4. Offensive security / penetration testing background (e.g., red team, bug bounty, exploit development).