AI Runtime & Trust Security Advisor II

Fiserv•Alpharetta, NJ

2d•Onsite

About The Position

The AI Runtime & Trust Security Advisor II helps lead the security of AI in production — both what deployed AI does and what it is allowed to do — detecting AI-specific threats such as prompt injection, jailbreaks, and inference-time data exfiltration while helping govern identity, authorization, and secrets for AI systems and agents. The role leverages AI protection tooling along with building detection in-house: commercial AI observability and guardrail platforms (e.g., Arize, Fiddler), data security posture management for AI (e.g., Varonis), and the enterprise SIEM/SOAR, IAM, and secrets stack. Sitting within Cybersecurity, this role helps lead a cross-functional model that connects Cyber's services with the SOC, IAM, Data Governance, and platform teams as Fiserv scales AI in production.

Requirements

7+ years in security engineering spanning at least two of: AI/ML runtime security, security operations and detection engineering, and identity / trust engineering.
Hands-on familiarity with AI attack patterns: prompt injection, jailbreaks, model inversion, and inference-time data exfiltration.
Experience operating commercial AI observability and/or guardrail platforms (Arize, Fiddler, or equivalent) and SIEM/SOAR (Splunk, Microsoft Sentinel, Chronicle, or equivalent).
Direct experience securing non-human and service identities, including workload identity standards (SPIFFE/SPIRE, OIDC) and authorization policy engines (Open Policy Agent, Cedar).
Working knowledge of data security posture management (Varonis, Cyera, BigID, or equivalent) and secrets management (HashiCorp Vault, AWS Secrets Manager, Azure Key Vault).
Comfort partnering across the SOC, IAM, Data Governance, and platform teams, leading shared outcomes through influence.
Bachelor's degree in Computer Science, Cybersecurity, or a related field, and/or equivalent work experience.

Nice To Haves

Experience in financial services or other regulated industries with AI-specific compliance, privileged access management, or non-human identity governance at scale.
Hands-on experience with LLM red teaming or adversarial ML testing.
Experience designing authorization models for multi-agent AI systems — delegation chains, capability scoping, and time-bounded access grants.
Certification in CISSP, CISM, CCSP, or equivalent; security operations or IAM certifications.
Familiarity with AI network egress controls (tool-call boundaries, model API access) and zero trust applied to machine identities.

Responsibilities

Lead operationalizing runtime monitoring, guardrails, and model-integrity checks for production AI workloads using commercial AI observability and guardrail platforms — detecting prompt injection, jailbreaks, model abuse, drift, and anomalous inference, and tuning thresholds and alert quality.
Serve as the primary SOC liaison for AI security events — providing platform context, triage support, and expertise on AI-specific attack patterns general SOC analysts may not recognize — and help maintain AI-specific incident response playbooks.
Help lead the enterprise authorization model for agentic AI — partnering with IAM and the AI Center of Excellence to define what agents may do, on behalf of whom, with what tool scope and duration — and help extend the identity fabric to AI agents as first-class principals (workload identity standards such as SPIFFE/SPIRE, OIDC).
Operate identity anomaly detection and policy enforcement for AI workloads and agents, adjudicating scope violations and out-of-scope tool calls.
Help lead data security posture for AI in partnership with Data Governance — operating DSPM tooling (e.g., Varonis, Cyera, or equivalent) to detect and reduce over-permissioned data exposure to copilots and agents, and responding to AI data-exposure findings.
Partner with the Cloud and platform teams to define and operate secrets hygiene and network egress controls for AI workloads, particularly agent tool-call boundaries and model API access.
Adjudicate novel authorization requests and escalations that require human judgment on risk tolerance, and lay the groundwork for a future automation layer as the program matures.