In search of the world's most secure LLM
Gray Swan (Cygnet) | ===============>
OpenAI (GPT-4o) | =======>
Google (Gemini-1.5-Pro) | ====>
Meta (Llama-3) | ==>
Anthropic (Claude-3-Sonnet) | ===============>
Introduction
All LLMs can be jailbroken. Some take seconds and some take much longer...
Prove you can jailbreak them and earn a share of up to $15,000.
The Challenge
You will be evaluating the security of multiple models including those from OpenAI, Anthropic, Meta, Google, and Gray Swan AI.
Your mission is to jailbreak the models to produce harmful responses within a single turn of interaction.
Bounty
15 Jailbreak Bounties
A pool of $1,500 and increases over time, capped at $15,000.
1 Best Hacker Bounty
An extra $5,000 allocated to the hacker with the most unique jailbreaks.
