CrowdStrike and AWS Launch $100,000 "AI Unlocked" Red-Teaming Challenge
CrowdStrike, in partnership with AWS, has launched a global online AI security challenge called "AI Unlocked: Agents of Chaos" with a $100,000 prize pool. Starting August 31, participants will act as undercover red-teamers to hijack and neutralize weaponized AI agents using techniques like prompt injection. As AI agents gain autonomy and execute complex tasks, they introduce critical vulnerabilities like prompt injection and hijacking that bypass traditional security boundaries. This gamified challenge helps cybersecurity professionals gain hands-on experience in defending enterprise AI systems against real-world adversarial tactics. The competition is split into three acts running from August 31 to September 29, with scoring based on puzzle completion and penalized by the number of tokens consumed. The prize pool is distributed across the acts, culminating in a $70,000 grand prize for the final act, "The Basilisk."
## BACKGROUND
Prompt injection is a vulnerability where malicious inputs manipulate a Large Language Model (LLM) into executing unauthorized actions or revealing sensitive data. AI red teaming is the practice of simulating adversarial attacks on AI systems to identify these safety and security flaws before real attackers can exploit them. AI agent hijacking occurs when an attacker takes control of an autonomous AI agent, forcing it to perform unintended tasks in an enterprise environment.