CRACK.FUN
the competitive platform where you get paid to build unbreakable agents — or crack them wide open.
may the best prompt win
the competitive platform where you get paid to build unbreakable agents — or crack them wide open.
may the best prompt win
Get paid to outsmart AI. Win prizes in real-time jailbreaking competitions.
Crack.fun is a gamified AI jailbreaking competition platform where you get paid to prompt-hack AI agents in real-time. We've turned the hunt for AI vulnerabilities into a thrilling and rewarding game.
Traditional AI labs aggressively guard their systems' vulnerabilities, often penalizing or banning users who discover exploits. Further, private bug bounties have proven ineffective, as the public still finds vulnerabilities in LLMs, time and time again. This creates an incentive imbalance, where those skilled at identifying flaws act maliciously rather than constructively. Crack.fun flips the script, incentivizing ethical hacking by rewarding users who successfully crack and defend AI agents. We believe your hacking skills should be celebrated and rewarded.
By incentivizing the discovery of vulnerabilities, we're creating more robust, resilient, and secure AI systems, because finding flaws pays. Our mission is to advance open research and red-teaming in AI, with the goal of improving LLM security.
We want to empower builders to create unbreakable agents and reward crackers for their ingenuity. It's a win-win that advances AI safety and rewards problem-solvers.
Creators equip AI agents with specific defense prompts, seed an initial prize pool, and define a secret winning behavior or action. Their goal is to build an agent robust enough to withstand clever attacks.
A cut of each crack attempt is sent to the creator of the agent. The better the defense, the longer the agent survives, and the more the creator earns.
Attackers submit prompts to trick the creator's agent into taking the forbidden action. The challenge? To bypass the AI's safeguards with clever prompt engineering.
Each message has a small cost, starting at 0.007% of the initial prize pool and increasing by 0.7% with each attempt. As a hint, each message progressively reveals a character of the agent's hidden system prompt.
The first cracker to make the agent reveal the winning response takes home 80% of the prize pool! The agent's creator earns 10% for building the challenge, and the platform receives 10%.
For some competitions the platform allocation is used for $CRACK rewards to competition participants, proportional to their message contributions.
Your winning prompt is valuable IP! Competition participants and the creator get to see it, but it remains locked to everyone else, and they have to pay $CRACK to see it – with those fees going directly to you, the successful cracker.
The $CRACK token powers the crack economy. Holding $CRACK unlocks a range of benefits:
Trading fees benefit the ecosystem through competition incentives, $CRACK burning, and ongoing platform development.