Immersive, which dropped the Labs from its name, is a Bristol-founded security training company that mostly sells to large enterprises. Its main platform, Immersive One, has a big lab catalogue that now includes a Building with AI category and labs on the OWASP Top 10 for LLMs and on NVIDIA NeMo Guardrails. Individuals cannot just sign up and pay for it, though. What the public can use for free comes down to two things. The first is Cyber Million.
It is a free version of the platform for anyone over 16, with no application process and no time limit. There are two paths: Cyber Fundamentals, with 50+ exercises on ethics, basics and terminology, and Defensive Security Operations, with 125+ exercises. Finishing the second path lets you apply for entry-level SOC roles posted by employer partners. Accenture and Arctic Wolf have both been partners.
The labs are short and practical, and the career link is the best part. Most free training ends with a certificate. This one ends with a list of actual jobs. The second is the prompt injection challenge. You talk to a chatbot that has been told a password and must not reveal it, and over 10 levels the defences get tougher.
Immersive used it to publish research showing that most players got the bot to leak the password. It is a decent hour of fun and makes the point that system prompts are not security controls. But it is basically the same idea as Lakera's Gandalf, and Gandalf has more levels, more variants and more explanation. If you only do one, do Gandalf. My main frustration is that Immersive's useful AI material, the guardrail and LLM Top 10 labs, is enterprise-only.
If your employer has an Immersive contract, ask for access, because those labs look like some of the more practical AI security training around. If not, you are left with a free SOC path that has nothing to do with AI and a short password game. That still makes it worth a sign-up. Cyber Million is a good free route into defensive work, and those skills matter more now that AI tools are writing first drafts of incident reports that someone has to check.