22. AI Security Engineer
Evidence label: Role family. Cisco, Apple, GuidePoint Security, Marvell, Prologis, Société Générale, and other employers use the exact title or a seniority-qualified variant in current official postings.
What they do
Protect AI applications, models, data paths, agents, and supporting platforms against misuse and attack. AI Security Engineers combine application, cloud, identity, data, and adversarial-ML security rather than treating prompt injection as the whole problem.
Apple's AI Security Engineer, Red Team focuses on deep technical reviews. Marvell's AI Security Engineer owns enterprise controls, runtime protection, identity, telemetry, and remediation. Société Générale's Lead AI Security Engineer combines AI red teaming with secure software engineering.
Responsibilities
- Threat-model model APIs, RAG, agents, tools, data flows, identities, and third-party components
- Test prompt injection, data exfiltration, poisoning, insecure tool use, privilege escalation, and model or supply-chain abuse
- Build preventive and detective controls for inputs, outputs, runtime actions, data access, and credentials
- Integrate AI findings into AppSec, cloud security, IAM, detection, incident response, and vulnerability-management workflows
- Validate mitigations with repeatable tests and retain evidence for risk owners
Distinction from adjacent roles
| Role | Primary question |
|---|---|
| AI Security Engineer | How can an attacker, compromised component, or unsafe integration violate a security property? |
| AI Safety Engineer | Which model or system behaviors create unacceptable harm or misuse risk? |
| AI Governance Engineer | How are policies, approvals, inventories, evidence, and regulatory controls encoded and operated? |
Entry paths
Application security, cloud security, security architecture, adversarial ML, IAM, or platform security with hands-on AI application knowledge.