CyberSN counted 2,409 U.S. AI Security Engineer job postings from January through September 2026, up 268.3% from the same months in 2025. Across cybersecurity as a whole, it counted 235,786 postings, down 26.4%. These figures describe advertisements, not hires or employed people. The role itself covers hands-on defense of AI models and applications, and the requirements vary widely by employer and seniority.
What CyberSN’s job counts measure
CyberSN’s figures cover U.S. postings from January through September 2026. The company says its data came through JobSpikr, were deduplicated, and were classified by the work requested. It pulled the data on October 7, 2026.
| Posting series | U.S. postings, January–September 2026 | Change from January–September 2025 |
| Cybersecurity overall | 235,786 | −26.4% |
| AI Security Engineer | 2,409 | +268.3% |
CyberSN treats AI Security Engineer as a role category within the broader dataset, so the two counts are not additive. A separate measure found AI or machine-learning terms in 14.6% of postings over the same nine months, compared with 15.1% a year earlier. That keyword share counts mentions across job descriptions; it is not a count of specialist AI-security roles.
What an AI security engineer does
The work is about protecting AI systems themselves. CyberSN describes assessing models and applications for risks such as prompt injection, jailbreaking, data poisoning, model extraction, and sensitive information leaking through outputs. Engineers may protect training data, model weights, inference endpoints, and third-party models or datasets, then test systems for weaknesses and implement safeguards such as input and output filters, guardrails, and access controls.
That work can involve machine-learning frameworks such as PyTorch and TensorFlow, model registries, MLOps pipelines, cloud AI services, and vector databases. CyberSN also names MITRE ATLAS, the OWASP Top 10 for LLM Applications, and the NIST AI Risk Management Framework as relevant guidance. The role guide classifies AI Security Engineer under defense and at the mid-career level.
How the work and requirements differ by employer
The title can describe very different jobs. Microsoft’s example centers on controlled evaluations of AI models and agents: designing tests, setting baselines, collecting telemetry, and working in cyber ranges with containment controls. MITRE focuses on adversarial AI analysis and defenses for government systems. DigitalOcean’s principal-level role combines product security and adversarial testing with lifecycle threat modeling, governance, and communication with senior leaders.
| Employer and role level | Main work | Stated experience and conditions | Advertised annual compensation |
| Microsoft, Security Research IC3 | Controlled evaluations of AI models and agents, with cyber-range testing and containment | A relevant master’s degree plus at least one year of qualifying experience, a relevant bachelor’s plus at least two years, or equivalent experience | Base pay: $102,100–$202,200 nationwide; $133,800–$219,200 in the San Francisco Bay Area and New York City metro areas |
| Microsoft, Security Research IC4 | Controlled evaluations of AI models and agents, with cyber-range testing and containment | The same listed degree-and-experience routes or equivalent experience | Base pay: $119,800–$234,700 nationwide; $160,200–$261,000 in the San Francisco Bay Area and New York City metro areas |
| MITRE, AI Security Engineer | Adversarial AI analysis, mitigations, and AI integration for government systems | Bachelor’s degree plus at least two years of AI experience, an advanced-degree and relevant-experience route, or an equivalent combination; U.S. citizenship for clearance eligibility, a TS/SCI clearance with CI poly within one year, and at least three onsite days per week | $121,600–$182,400 per year; listed midpoint $152,000 |
| DigitalOcean, Principal Engineer, AI/ML Security | Product security, adversarial testing, threat modeling, governance, and executive communication | 10+ years in cybersecurity, including 4–5 years in AI/ML security-related work; the listing describes a remote role | $230,400–$288,000 per year |
The Microsoft ranges are base pay, with separate figures for the listed metropolitan areas. MITRE’s clearance and onsite terms are specific to its position; they are not general requirements for AI-security work. DigitalOcean’s experience threshold applies to a principal role, not to every job with AI security in its title.
What the pay examples tell candidates
The advertised ranges are useful reference points for the named positions, but they span different employers, levels, and conditions. They do not establish a U.S. national median or average for AI security engineers. Microsoft’s listing separates IC3 and IC4 compensation and gives higher ranges for two metropolitan areas; MITRE’s role has its own clearance and onsite requirements, while DigitalOcean’s figure is for a principal-level position.
Experience paths and skills
CyberSN says candidates need working knowledge of how machine-learning systems are built and deployed, but do not have to be research scientists. It describes paths into the field from security engineering, application security, or machine-learning engineering. The Microsoft listing accepts equivalent experience in place of specified degree routes, and MITRE also lists an equivalent combination of education and work experience. Their experience, clearance, and onsite conditions still differ, so applicants should compare the full requirements of each role rather than treat one pathway as universal.
CyberSN names AIGP, AAISM, and CISSP as credentials associated with the field, while saying that no certification is universally required. It also says hands-on work carries at least as much weight in its hiring conversations. In his analysis published April 8, 2026, Josh Madakor reported that machine-learning skills appeared in 57% and AI governance in 50% of his self-compiled sample of 100 cybersecurity-and-AI listings. Those percentages describe his sample, not the national posting dataset.
