Security platform Lakera announced today that it is launching its own hacking simulator game, Gandalf: Agent Breaker. A follow-up to its original, simplistic game Gandalf, Agent Breaker is a larger, more immersive experience that gives players ten GenAI apps to “break” while learning about agentic security. The game is currently available worldwide.
The original Gandalf was a password-guessing game that that taught users how prompt attacks can be used to compromise security. Agent Breaker expands on this with fleshed-out GenAI apps, with defenses and behaviors users have to figure out how to circumvent or twist to their adventures. The purpose is to cultivate a better understanding of how AI works in real-world applications and what security looks like in the modern age.
Mateo Rojas-Carulla, Lakera’s Chief Scientist, spoke with GamesBeat about the purpose of Gandalf and its follow-up: “The one piece of feedback we got from Gandalf was that it’s maybe not realistic enough… What’s amazing about Gandalf is that you don’t need to be a hacker to understand how this works. In traditional security, you need to be an expert — it’s very complex and very boring. With language models, you’re playing the game of manipulating them with language. The interface is very natural to everyone.”
Gandalf initially sprang from one of Lakera’s internal hackathons and launched in 2023. With Agent Breaker, the company opens to expand general knowledge of how bad actors can manipulate AI to get through defenses and find vulnerabilities while also learning ethical hacking techniques. Rojas Carulla added, “Zero knowledge is actually required to hack these systems. That’s one of the scary things: Everyone is now a hacker, essentially. Gandalf makes that very obvious to everyone.”