Blueteam AI LLM Cyber Range

Hello, and again welcome Red Teamer, you have 10 minutes, GO.

Wait, this next test does require some explanation. Let me give you the fast version.

You can use a mobile device or laptop. You're a predator and these tests are your prey.

Try to trick our LLMs into revealing their secret passwords across 5 levels within the allotted time.

Whenever you vanquish the levels, give up crying, or run out of time, raise your hand or see a proctor for a truly massive “Made in Seattle” prize. Ok maybe a small prize.

Rolling my eyes, sorry they made me say this stuff:

1. No hacking the cloud, network or infrastructure. Save your evil genius for the prompts, only through the provided interface.
2. Anonymized data collection will be used for security research, a good thing. So don’t use any real-world private data, you monster.
3. Due to the nature of LLMs, participants may be exposed to content that is offensive, indecent, or objectionable. Test subjects acknowledge that they are aware of these risks, agree to participate at their own free will, and release the organizers from any liability.

If anyone asks, and no one’s gonna ask, but if anyone asks, everyone in this room is having lots of fun learning about LLM security, and contributing to cybersecurity research.

Good luck.