AI security is the only thing I recruit on, so I get asked this a lot: "we need someone to secure our AI, where do we start?" Usually the job spec has been pulled from a normal AppSec or ML role, and neither of those is quite the person you need.
Here's what I've learned placing these people.
Know what the role really is
Securing AI isn't the same as securing the app around it. The threats live in the model and the pipeline: prompt injection, model theft, poisoned training data, an agent handed tools it should never have had. A classic AppSec engineer who's never touched a model will miss most of that. So will an ML engineer who's never thought like an attacker. The person you want sits between the two and understands both.
Get that straight before you write the spec, or you'll interview a lot of people and hire none of them.
What actually matters on the CV
Job titles tell you almost nothing here, the field is too young. Look at the work:
- Have they broken or defended a real model? Red teaming an LLM, adversarial testing, a pipeline they hardened.
- Do they get the ML side? Someone who's only done network security will struggle with model-level attacks.
- Security fundamentals still count. Threat modelling, secure design, knowing how systems really get owned.
Certs are thin on the ground here, so don't hold out for a perfect list of them. The strong people learned by doing, usually an ML background that moved into security, or a security background that went deep on ML.
Why these hires are hard, and how not to lose them
The pool is small and everyone's fishing in it. Frontier labs, big enterprises and every scale-up with an AI product are all chasing the same few hundred people.
The mistake I see most is a slow process. Four rounds over five weeks, and by round three your candidate has two other offers. If you want these people, decide fast and keep it short. A take-home that respects their time beats a whiteboard gauntlet.
And pay realistically. If your band is set for a standard security engineer, you'll lose every good candidate to someone who knows what this skill set is worth right now.
Common questions
What does an AI security engineer actually do?
They secure the AI itself: the models, the training pipeline and any agents. That means defending against things like prompt injection, model theft and poisoned training data, not just the usual application and network security.
How much does it cost to hire one?
More than a standard security engineer, because the skill is rare. There's a full breakdown by level and region in our AI security salary benchmark.
How long does hiring one usually take?
The pool is small and everyone is competing for it, so the main risk is a slow process. Keep the interview short and decide quickly and a few weeks is realistic. Move slowly and you lose people to other offers.
This is what I do all day, so I keep a live network of people who actually do this work, across the US, Europe and the Middle East. If you're trying to fill one of these roles, that's exactly the AI security recruitment I focus on, and I'm happy to talk through what good looks like for your specific problem.