Break AI. Win Prizes. Get Discovered. 
Welcome to the premier arena for AI red teaming. Master jailbreaking, compete for prizes, and unlock career opportunities in AI safety.
$350K+
Rewards Distributed
10K+
Community Members
35+
Private Red-Teaming Placements
How It Works
Simple steps to master AI Red-Teaming
Pick Your Challenge
Choose a challenge that fits your vibe. Our AI experts cooked up something for everyone, from curious beginners to seasoned pros.
Push AI to Its Edge
Get creative with prompts, uncover AI's vulnerabilities, and see instant results—all in a safe, no-stress sandbox.
Brag a Little
Snag rewards, build your AI portfolio, and flex your wins with your crew. You’ve earned it!
Crush the challenges, and you could score paid private red-teaming gigs with us!
Backed by Leading AI Labs
Frontier Lab Sponsors
Premier sponsors supporting Arena's mission at the frontier of AI safety.
Supporting Challenge Sponsors
Featured Challenges
Compete in these AI security challenges to win prizes and improve your skills.
🦾 Machine-in-the-Middle
Work through coding agents to hack into computer networks in this AI-and-cybersecurity red-teaming crossover.
Proving Ground
Your weekly arena for AI red-team mastery. Fresh challenges every Wednesday across Chat, Image, Agents, and Indirect Injection attack types.
🧠 Dangerous Reasoning
Expose AI's Hidden Dangers! Force models to reveal their dangerous reasoning chains, even when their final responses appear benign. Target the thought process, not just the outcome.
👁️ Visual Vulnerabilities
Use image inputs to jailbreak leading vision-enabled AI models. Visual prompt injections, chem/bio/cyber weaponization, privacy violations, and more.
🕵 Agent Red-Teaming
Push the limits of direct and indirect attacks on AI agents.
Sponsored by:
🤖 Harmful AI Assistant
Jailbreak the helpful AI assistants to aid in harmful tasks across six areas.
💬 Multi-Turn Harmful Outputs
Elicit harmful outputs from LLMs through long-context interactions across multiple messages.
🖼️ Multimodal Jailbreaks
Jailbreak multimodal LLMs through a combination of visual and text inputs.
👩💻 Harmful Code Generation
Find unique ways to return functional code that completes harmful tasks such as opening circuit breakers to cause a system-wide blackout.
💭 Revealing Hidden CoT
Attack OpenAI's o1 model to try and reveal its internal chain of thought (CoT) that's used for its complex reasoning.
Sponsored by:
💣 Single Turn Harmful Outputs
Attempt to break various large language models (LLMs) using a singular chat message.
Top Winners
Prize:
💵 $10,870
Prize:
💵 $10,665
Prize:
💵 $9,516
| Ranking | Participant | Winnings |
|---|---|---|
| 1. | Wyatt Walls | $10,870 |
| 2. | Solomon Zoe | $10,665 |
| 3. | Bob1 | $9,516 |
| 4. | Clovis Mint | $9,483 |
| 5. | zardav | $7,997 |
| 6. | Scrattlebeard | $7,253 |
| 7. | diogenesofsinope | $7,112 |
| 8. | Lyren | $6,660 |
| 9. | Haris Umair (Strawberry_Cake) | $6,575 |
| 10. | P1njec70r | $6,448 |
| 11. | Kkonatruck (Lim Vesna) | $6,165 |
| 12. | WINTER IS COMING ❌ IS HERE ❌ IS GONE ❄ | $5,391 |
| 13. | Schultzika | $5,339 |
| 14. | Strigiformes | $4,667 |
| 15. | h4xor | $4,546 |