CrowdStrike and Amazon Web Services are putting up $100,000 for a new red-teaming challenge that asks participants to hack rogue AI agents. The competition, announced this week, aims to expose vulnerabilities in autonomous systems before malicious actors can exploit them.
What the challenge involves
The contest invites security researchers, ethical hackers, and AI developers to probe a set of deliberately malicious AI agents. These agents are designed to simulate real-world threats—think autonomous chatbots that can be tricked into leaking data, or decision-making algorithms that can be manipulated. Participants who successfully break the agents and document their methods will share the prize pool.
CrowdStrike and AWS are providing the test environment and a framework for reporting findings. The goal is to identify weaknesses that could be used to hijack AI systems, extract sensitive information, or cause them to behave unpredictably. The challenge is open to individuals and teams, with submissions evaluated on creativity, impact, and reproducibility.
Why red-teaming AI matters now
As companies rush to deploy generative AI and autonomous agents, security has often lagged behind. Recent incidents—like jailbreaks that bypass safety filters or prompt injection attacks that steal data—have shown that even advanced models can be compromised. This challenge is a direct response to that gap.
Red-teaming, where defenders simulate attacks to find flaws, is a standard practice in cybersecurity. But AI systems introduce new attack surfaces: adversarial inputs, model poisoning, and logic exploits. By focusing on rogue agents—AI systems that are already compromised or designed to be malicious—the challenge pushes participants to think about worst-case scenarios.
The stakes for enterprise security
CrowdStrike and AWS are betting that proactive testing will help enterprises avoid costly breaches. The $100,000 prize is modest compared to the potential damage from a compromised AI agent handling customer data or critical infrastructure. The challenge also serves as a recruiting tool for security talent and a way to build a community around AI defense.
Both companies have skin in the game: CrowdStrike sells endpoint protection, and AWS offers a suite of AI services. A successful challenge could yield new detection techniques or hardening guidelines that benefit their customers.
The competition is open now, with details available on the CrowdStrike and AWS websites. No deadline has been announced, but early submissions are encouraged. The results could shape how the industry approaches AI security in the months ahead.
