AI Security Research Engineer

What You'll Do

Leads strategic AI security initiatives and roadmaps with manageable-to-significant complexity (e.g., six-month to one-year plans) across the team's three pillars: securing AI-assisted development, securing AI in products, and securing AI in how we run the business.

Influences product organizations and partners across S&TO, setting the security bar while partners build and run.

Applies deep expertise in AI/ML security and research methods (adversarial machine learning, prompt injection, model extraction/inversion, membership inference, data poisoning, and excessive agency) while weighing ethical and responsible-AI considerations.

Sets AI red-team methodology and evaluation strategy; advances eval harnesses, LLM-as-a-judge approaches, and ground-truth/silver-dataset generation so conformance and agent behavior are measured with evidence rather than described manually.

Owes secure AI architecture patterns and prohibited design patterns for AI-native features (RAG, tool use, multi-agent workflows, MCP (Model Context Protocol) integrations), including prompts, context, retrieval, model selection, and output handling.

Authors and shepherds AI security standards and security requirements; defines clear release criteria for AI systems.

Drives the model and data supply-chain security strategy, including model provenance and data protection requirements such as classification, minimization, residency, and sensitive data handling.

Defines agent identity, human review, approval, and accountability controls so autonomous agents and AI-assisted outcomes are authenticated, bounded, and accountable.

Owns reusable platform contributions and architecture (security tooling and agentic workflows) that scale enablement across engineering rather than blocking it.

Monitors and evaluates emerging AI technologies, attacker techniques, and adversarial research; develops hypotheses, designs experiments and prototypes, and turns discovery into shipped controls.

Synthesizes findings into triangulated insights and meta-analyses that impact product and security decisions across multiple teams and AI initiatives.

Independently develops and delivers presentations; leads cross-functional working sessions with diverse audiences and creates clear ownership models with partner teams.

Led creation of research artifacts (threat research, patterns, proposals, patents) of scientific quality and rigor; shares in company and industry forums.

Serves as the AI security technical authority within the team; mentors peers and security champions across engineering through reusable patterns, training, and office hours.

Minimum Qualifications

• Bachelor's + 7 years of related experience, or Master's + 4 years of related experience, or PhD + 1 year of related experience.
• Demonstrated depth in AI/ML security and hands-on security engineering, with a track record of leading initiatives and influencing cross-functional teams.
• Proficiency in Python; ability to read and review code fluently across multiple languages.

Preferred Qualifications

• Proven leadership of AI red-team engagements and evaluation programs; familiarity with Cisco AI Defense (Cisco's AI security solution) or an open-source framework such as NVIDIA's Garak.
• Deep command of LLM and agentic-system security: prompt injection (direct and indirect), jailbreaking, insecure output handling, RAG hardening, tool-call governance, and multi-agent trust boundaries.
• Experience authoring security standards or baselines and embedding them into a secure SDLC and CI/CD.
• Authority with OWASP LLM/Agentic Top 10, MITRE ATLAS, NIST AI RMF, ISO/IEC 42001, and the EU AI Act, with the judgment to apply them proportionately by risk tier.
• Experience shipping or integrating agentic and MCP (Model Context Protocol) systems and AI development tooling (Cursor, Claude Code, Copilot).
• Experience with Go, Rust, or C/C++ is a plus, with C/C++ valuable for Cisco's embedded and networking products.
• Recognized external contributions, such as published threat research, conference talks, open-source security tooling, or patents.