Other12 modules, self-paced, plus a 7-day exam·Subscription (Cubes from about $8; Silver Annual recommended for the certification)

Hack The Box Academy AI Red Teamer Path and HTB COAE

4.4

The most complete hands-on AI red teaming curriculum I have found. It goes well past prompt injection into evasion, poisoning and privacy attacks on models, and the certification makes you write a real report. The pricing model is confusing and it assumes you already know offensive security.

What We Liked

Covers the full range: prompt injection, adversarial examples, data poisoning, model privacy, supply chain and deployment attacks
Built with Google and mapped to SAIF, so the structure follows a recognised framework
Real labs rather than slides, in the usual HTB style
COAE exam is 7 days and needs a professional written report, which is closer to real work than multiple choice
Useful for ML engineers who want to see how their models get attacked

What Could Be Better

Cubes, tiers and subscriptions make it hard to work out what you will actually pay
Assumes solid Python and some offensive security background, not a beginner path
The certification is new (April 2026), so hiring managers may not recognise it yet
Search interest in the path is small, which says something about current demand for the role

Detailed review

Hack The Box made its name with vulnerable machines for penetration testers, and HTB Academy is its structured course side. The AI Red Teamer job role path was built with Google and follows Google's Secure AI Framework. It has 12 modules covering the fundamentals of how models work, prompt injection and jailbreaks, attacks on AI applications, adversarial examples and evasion, data poisoning, model and data privacy attacks such as membership inference, and supply chain and deployment risks. The range is the main reason I rate it. Most AI security training stops at prompt injection because that is easy to demo in a browser.

This path spends real time on attacks against the model itself, which means you write code to craft evasion inputs and poison training sets, and you see how little it takes. For ML engineers that is sobering in a useful way, and I would argue they get more from it than pentesters do. The path leads to the HTB Certified Offensive AI Expert, or COAE, launched on 2 April 2026. The exam runs for seven days and you finish by submitting a commercial grade technical report. I like that design.

It tests whether you can find and explain problems to a client, which is the job, rather than whether you remember definitions. The weak spot is pricing. HTB Academy runs on Cubes, a credit system with a small free starting balance, alongside tiered subscriptions, and HTB recommends the Silver Annual plan for the certification without the announcement stating the price. Work out the total before you commit. The other caveat is maturity.

The credential is months old, and dataforseo shows only a trickle of searches for the path, so do not expect recruiters to ask for it yet. My take: excellent content for a niche that is growing slowly. If you already do offensive security or you build models and want to understand attacks, it is the best structured option I have seen.

[ final ]

The verdict.

The best choice if you want to red team models themselves, not just the chatbot in front of them. Do PortSwigger's free LLM labs first, then pay for this if the topic holds your interest.