Ask most security leaders whether their team is ready for AI risk and you’ll get some variation of a yes. Ask them to specify what exactly they’re ready for, and the answer will get murky. The industry has been nodding along to "AI ready" for some time, without agreement on a definition.
AI-readiness is more than having deployed a tool. It means being clear on the problem you expect AI to solve, where the limitations of its remit are, and understanding specifically what risk you are taking on by using AI in a given part of your business. Crucially, it’s also knowing whether your team actually has the skill to identify and manage that risk. Too few organizations have not done the work needed to clearly answer on all these points. In fact, they need to revisit these questions continuously as the answers evolve.
AI adoption has outpaced the problem definition
We spoke with security leaders across healthcare, finance, automotive, manufacturing and insurance while building our new AI security path, and one theme came back again and again: AI systems are already running in production, handling sensitive data, in every one of those industries, and much of that happened without anyone stopping to define what problem the AI was actually solving or what new exposure it was creating in return.
Teams are being asked to protect systems they do not understand end to end, so what usually gets deployed in response is a patchwork of small, disconnected controls rather than a coherent framework built around an actual understanding of the risk. That is the same pattern that shows up whenever a tool gets adopted ahead of a clear problem statement.
Supply chain risk came up unprompted in nearly every conversation TryHackMe had, and it is a clean example of the same challenge: organizations adopting models and dependencies without first asking what they would need to verify or monitor once those components were inside their environment.
Training built on the wrong assumption doesn’t protect
Part of why this keeps happening is that most available training reinforces the assumption that AI risk is a known, static thing you can learn once and check off. Watching a lecture on prompt injection can make the concept click instantly, and then fall apart the moment someone sits in front of a real system with actual guardrails and tries to extract the system prompt themselves.
Teams that can pass a knowledge check often cannot reason through a genuinely novel problem, which suggests the theory never really transferred into judgment. Much of the training on the market only simulates the AI itself, which makes it deterministic: the same input always produces the same output.
Real AI systems do not behave that way. Two identical prompts can return two different responses, and that unpredictability is the actual attack surface defenders are up against. If your training does not reproduce that unpredictability, it cannot teach anyone to define the real problem they are facing, only a simplified stand-in for it.
Readiness means defining the scope of AI’s remit
Genuine AI readiness looks a lot like the discipline any good problem definition requires, applied specifically to security. It means being able to say, for a given system, exactly what you are trying to protect against, who owns that risk, and where the human needs to stay in the loop regardless of how convincing the AI's output looks.
That accountability does not sit exclusively with the security team. Every engineer writing code with an LLM, every team pulling a model from a repository, every workflow with AI stitched into it is expanding the exposure, and a manager who pushes AI adoption for the sake of productivity is also, whether they realize it or not, taking on responsibility for what happens if that usage was never scoped properly. A security incident tied to ungoverned AI use rarely stays contained to the security function. It carries consequences for brand trust, regulatory standing, and revenue, and it traces back to the point where nobody paused to ask what specifically was being adopted, and why.
Building capability across teams
Getting your organization ready for safe AI implementation does not start with more tooling. In a security context, it means hands-on practice against real, non-deterministic AI systems rather than simulations, coverage broad enough to span the full threat landscape rather than one narrow slice of it, and honest engagement with supply chain risks that most training skips past entirely. Things like backdoored tokenizers, poisoned checkpoints, and typosquat dependencies hiding in a model repository. It also has to reach past the security team itself, since developers, DevOps, and engineers are already on the front line of AI usage whether or not their job title reflects that.
AI is an opportunity to reframe the role of the security team
AI is an opportunity for the security team to break out of its reputation as the as the team that says no, and claim a strategic role in guiding the business through implementation and adoption.
Security has spent a long time earning a reputation as a blocker, preventing or stalling decisions made elsewhere in the business, often without its involvement. AI adoption is happening at a pace and with a level of executive pressure that makes that old posture untenable. Saying no to a board level mandate does not work, and it should not be the goal. What works, and what actually builds influence, is being the voice in the room asking the question necessary to prevent legal liability, lost reputation and reputational damage that can’t be easily undone. Far from obstruction, this is strategic leadership that can align a group of stakeholders feeling the rushed to achieve vaguely defined goals.
To make that shift, teams need to push to be looped in while decisions are still being shaped, proving they can move at the speed of the business. Success will be more than a successful implementation of new technology, it’ll be a redefinition of the organization’s relationship to security as a discipline and as an organizational function.
That redefinition includes taking on a greater role as stewards of understanding across the org. With the vulnerabilities AI usage creates, security needs to become a shared language, shared goal and shared responsibility in every department. Superficial yearly awareness training isn’t enough to manage threats in a post-AI world. The cyber team can lead this change, building strategic allies along the way.
Joanna Duffy