Staff AI Security Specialist

Job Description

Our team secures the AI in ServiceNow's products and platform from adversaries. As AI agents gain the ability to reason, retrieve context, and take real action on behalf of users, they become a new and largely unmapped attack surface - and the controls that protect them mostly don't exist yet. We build them.

We're looking for a Staff AI Security Specialist who sits at the intersection of AI and Security and is happiest building. This is an applied role: you'll spend your time reasoning about how agentic systems break and then writing the code that stops it, from prototype to shipped control in collaboration with Security Engineers.

What you get to do in this role:

• Analyse the security of AI agent systems - how the harness assembles context and drives the tool-calling loop, how agents plan and delegate, what they persist to memory, and how all of it responds to untrusted input - and find where an adversary can subvert them
• Design and build security controls for agentic systems: static analysis of agent and tool configurations, runtime policy enforcement, behavioural detection, and mitigations for tool and goal hijacking
• Build working prototypes end to end, then partner with engineering to harden the ones that prove out
• Threat model new AI architectures and features before they ship, and turn findings into concrete controls rather than a list of risks
• Red-team our own agents and our own defenses: indirect prompt injection, tool abuse, privilege escalation across agent boundaries, data exfiltration through model and tool channels
• Build repeatable evaluations and benchmarks so we can measure whether a control actually holds, and prove it as models and products change underneath us
• Track offensive and defensive AI security research and judge quickly what's real, what's noise, and what should change our roadmap
• Advise product and platform teams building agentic features on secure design, and raise the team's collective depth in AI security
• Publish what's worth publishing - internal research, patents, and external talks

Qualifications

To be successful in this role, you have:

• Experience in leveraging the integration of AI capabilities into work processes, decision-making, or problem-solving. This may include using AI-powered tools, automating workflows, analysing AI-driven insights, or exploring AI's potential impact on the function or industry.
• 6+ years of combined experience in security, software engineering, or applied research, including 2+ years working hands-on with AI/ML or LLM-based systems, is required
• Strong working knowledge of the agent harness and its security-relevant seams: system prompt and context construction, the tool-calling loop, context window management and compaction, sub-agent delegation, sandboxing and execution boundaries, is required
• Working knowledge of agent memory and retrieval systems - what persists, who can write to it, and how it is read back - along with orchestration frameworks and tool protocols such as MCP, and the failure modes each introduces, is required
• Demonstrated depth in at least one security domain (application security, offensive security, authorisation and identity, or detection engineering), with the instinct to reason in terms of attacker capability, preconditions, and attack chains rather than checklists, is required
• Practical understanding of the AI-specific threat landscape - prompt, tool, and goal hijacking, indirect injection through retrieved content, memory poisoning and adversarial instruction persistence across sessions, insecure agent configurations, cross-session and cross-tenant data leakage, OWASP Top 10 for LLM Applications, MITRE ATLAS - is required
• Proficiency in Python, sufficient to independently build and ship working systems, is required
• Ability to take an ambiguous problem statement to a working artifact, own it end to end, and abandon it without ceremony when the evidence says it doesn't work, is required
• Ability to communicate technical concepts to both non-technical business users as well as technical stakeholders is required

Preferred qualifications:

• Experience building systems that make and enforce security decisions at runtime
• Experience building static or dynamic analysis tooling
• Experience with AI red-teaming, adversarial ML, or evaluating models and guardrails at scale
• Published security or AI research, patents, CVEs, or meaningful open-source contributions