The Rise of the AI Security Engineer: A New Discipline for an AI-Native World

We are witnessing the birth of a new profession in the blend of security engineering and security operations, a discipline that didn't exist five years ago because the systems it protects didn't exist five years ago. As artificial intelligence moves from experimental to essential and agentic systems begin to perceive, reason, act, and learn autonomously, we need defenders who can operate at the same velocity.

I'm talking about the AI Security Engineer.

At Snyk's inaugural AI Security Summit in San Francisco this past October, I stood before 400 AI innovators and security professionals and made a prediction: within three years, every Fortune 500 company will have AI Security Engineers on staff. Not as a nice-to-have, but as a survival imperative. The response in the room told me I might be conservative.

The fundamental shift in AI Engineering

Traditional applications are deterministic: given the same input, they produce the same output and you can test, audit, and secure them using established methodologies. Agentic AI systems are different in that they are non-deterministic by design. In other words, they reason, adapt, and take actions in the world.

An LLM-powered application might generate different outputs each time it runs and an autonomous agent might take a sequence of actions that no human explicitly programmed. This dynamism is precisely what makes AI so powerful and precisely what breaks our traditional security models.

Consider this: Sam Altman recently acknowledged that AI models are now "so good at computer security they are beginning to find critical vulnerabilities." If AI can find vulnerabilities at machine speed, adversaries will exploit them at machine speed. Our defenses can no longer churn, and they can’t stall. Our defenses must operate at the same tempo.

The attack surface has expanded in dimensions we're still mapping. Prompt injection. Memory exploitation. Model poisoning. Agent hijacking. Supply chain attacks on training data. Model theft through inference queries. These aren't theoretical; they're happening now, and most organizations lack the visibility to even detect them. At Snyk, we’ve recognized this tectonic shift and have put forward Evo as the next evolutionary leap in security for AI-native software.

Traditional AppSec is table-stakes, but AI demands more

With decades spent in cybersecurity, I'll be direct: our existing frameworks weren't built for this. For example, traditional AppSec teams are trained to find code vulnerabilities, not adversarial inputs that manipulate model behavior. Network security teams monitor traffic patterns, not the subtle data exfiltration possible through carefully crafted prompts. Even our most sophisticated threat models assume a level of determinism that AI systems fundamentally lack.

The challenge isn't that our security professionals are unskilled. They are, in fact, extraordinary. The challenge is that AI-native systems present attack vectors that exist nowhere else in our technology stack:

• Adversarial inputs: Unlike SQL injection, which exploits code flaws, prompt injection exploits the model's intended behavior. The vulnerability isn't a bug; it's how the system works.

• Data and memory attacks: Agentic systems with persistent memory can be poisoned over time, with malicious instructions embedded in seemingly innocent interactions. RAG and indirect prompt injection exploit these underlying infrastructures.

• Model supply chain risk: When you integrate an open source model, a remote API-enabled model from untrusted and ungovernable parties, or a third-party MCP server, you're inheriting risk you can't inspect with traditional code analysis.

• Behavioral unpredictability: An agent that can "learn" the wrong things. Detecting when an AI system has been subtly compromised requires understanding not just its code, but its behavior over time.

This is why we need specialists; security practitioners whose primary mission is securing these AI-first and AI-native systems.

Defining the AI Security Engineer

So what does this role look like? Based on what we've learned, standing up Snyk's own AI security capabilities and from conversations with hundreds of organizations on the front lines, here's my view of the essential profile.

The AI Security Engineer operates at the intersection of three traditionally separate disciplines: platform security, AI/ML engineering, and threat intelligence. They are equally comfortable discussing gradient-based attacks with ML researchers and explaining model risk to the board.

The AI Security Engineer is an adaptive operative. The AI Security Engineer thrives in ambiguity, learns from every security incident, and assumes adversaries will move faster than static controls can keep pace. They embody what we call the Agentic OODA loop: Observe, Reason, Act, Learn. This means continuous, automated where possible, and human-supervised where necessary.

Practitioners of the AI Security Engineer are builders as much as defenders. The AI Security Engineer designs secure-by-default architectures, then thinks adversarially about how they might fail. They instrument the detection pipelines that can spot behavioral anomalies in AI systems. They create the tooling that doesn't exist yet because in a new field, the tools haven't been written.

Most importantly, they understand that AI security is not just technical; it's about trust, alignment, and ensuring that the systems we're building serve the purposes we intend, without being subverted by malicious actors or drifting into harmful behaviors.

A proposed role definition for the AI Security Engineer

For organizations looking to formalize this function, here's a condensed role specification:

The strategic imperative for AI security

Consider what's at play. AI systems are being deployed for fraud detection, clinical decision support, autonomous operations, customer interactions, and code generation. These are production systems with real-world impact. A compromised AI system doesn't just leak data; it makes wrong decisions at scale, potentially for extended periods before anyone notices.

The regulatory environment is evolving rapidly: The EU AI Act, industry-specific guidelines, and emerging liability frameworks. Organizations need practitioners who can translate these requirements into technical controls and demonstrate compliance to regulators and auditors.

And then there's the trust dimension. Your customers, partners, and employees need to know that the AI systems they're interacting with are trustworthy. That they haven't been poisoned, manipulated, or compromised. Building and maintaining that trust requires dedicated expertise.

This is why, at Snyk, we've made AI security a strategic priority. Our Evo platform is purpose-built to empower AI Security Engineers, providing the visibility, policy automation, and agentic security orchestration they need to defend AI-native applications across the entire development lifecycle. But tools alone aren't enough; the industry needs to build the human capability to wield them.

Are you heading to RSA Conference this March 2026? We invite you to join our Masterclass training for AI Security Engineer and receive a certificate of completion for various modules on AI-BOM, Red Teaming, MCP Security, Agent Skills security, among other labs:

AI adoption recommendations for organizations

If you're a CISO, CTO, or engineering leader, here's my guidance for building AI security capability:

• Start now, even if small. Don't wait until you have 50 AI applications in production. Identify one or two engineers with the right aptitude and begin developing the practice. The learning curve is steep, and starting early builds institutional knowledge.

• Invest in training. This is why Snyk launched the AI Security Engineer certification program alongside our AI Security Summit. The skills required don't exist in most security or engineering curricula today. Hands-on training on securing AI-generated code, adversarial testing, MCP security, and the OWASP Top 10 for GenAI, all of which are essential.

• Create the organizational home. AI security can't be orphaned between security and AI engineering teams. Define clear ownership, reporting lines, and cross-functional integration points. The most successful organizations I've seen treat AI security as a first-class discipline with its own mandate and metrics.

• Embrace agentic security. Just as your AI systems are becoming agentic, your security systems must follow. Manual review and static rules can't keep pace with the dynamism of AI applications. Invest in platforms that provide adaptive, automated security orchestration that can observe, reason, act, and learn alongside the systems they protect.

• Measure what matters. Mean time to detect and remediate AI-related incidents. Coverage of AI systems under a defined security posture (hint: start with AI-SPM). Automation ratio. And crucially: is your security system learning? Are you seeing fewer repeated incidents over time?

Looking ahead

I believe we're in the early chapters of a multi-decade transformation where AI systems will become more capable, more autonomous, more deeply embedded in critical infrastructure, and the attack surface will expand in ways we can't fully predict today. The adversaries, state actors, criminal organizations, and, yes, other AI systems will all become more sophisticated. In this future, AI Security Engineers won't be a specialized niche. They'll be as common and as essential as application and cloud security engineers are today. Every organization that builds or deploys AI will need them, and every security team will need this expertise embedded.

The good news is that we're seeing remarkable energy in this space. The sold-out AI Security Summit showed me a community that's hungry to learn, to share, to build. The practitioners entering this field bring creativity and adaptability that give me genuine optimism. The profession is being invented right now, the threat models are being written, the tools are being built, and the frameworks are emerging. If you're a security professional wondering whether to specialize in AI security, or an AI engineer curious about the security implications of what you're building, my message is simple: this is where the action is. This is the frontier.

At Snyk, we're committed to being your partner on this journey. From Snyk’s AI Security Platform to the free and accessible training we offer at Snyk Learn, to the AI Security Engineer community we're fostering, our mission is to help you secure the AI-native future. Because that future is already here. The question is whether you’ll defend it.

Discover why traditional security can’t keep pace with modern development—and what you must do to protect your software at machine speed. Download "The End of Human-Speed Security" to learn how to shift to automated, continuous defenses that keep your teams and code safe as systems evolve.