CrowdStrike and AWS launch $100K AI red-teaming challenge where players hack rogue agents

CrowdStrike and Amazon Web Services launched AI Unlocked: Agents of Chaos, a $100,000 prize pool competition running from August 31 to September 29, 2026, where participants use prompt injection to exploit autonomous AI agents in a fictional scenario. The challenge, announced on August 5, is based on a February 23, 2026 research study by Harvard and Stanford researchers that documented vulnerabilities including prompt-injection attacks and unauthorized data exfiltration. CrowdStrike CMO Jennifer Johnson said, "Prompt injection, agent hijacking, and rogue AI behavior aren’t theoretical anymore.

Via itpro.com CrowdStrike and AWS launch $100K AI red-teaming challenge where players hack rogue agents The Agents of Chaos competition pays participants to exploit autonomous AI vulnerabilities through prompt injection, with implications for how crypto and DeFi protocols secure their own AI integrations. CrowdStrike and Amazon Web Services want to pay you $100,000 to break their AI. The catch: you have to do it by manipulating autonomous agents inside a fictional shadow organization called Agents of Chaos. The challenge, officially titled AI Unlocked: Agents of Chaos, was announced on August 5 and runs from August 31 to September 29, 2026. Participants will use prompt injection techniques to hijack AI agents, essentially tricking autonomous systems into doing things they were never designed to do. How the challenge works The $100,000 prize pool is split across three escalating acts, each progressively harder and more lucrative. Act 1 runs from August 31 to September 7 and offers $10,000. Act 2 extends through September 14 with $20,000 on the line. Act 3, the final stretch from September 15 to 29, carries a $70,000 purse. Players enter a virtual scenario where they represent a fictional shadow organization. Their mission is to manipulate autonomous AI agents using techniques like prompt injection, a method where carefully crafted inputs override an agent’s original instructions. The competition builds on a research study also called Agents of Chaos, published on February 23, 2026. That study, conducted by researchers from Harvard and Stanford among other institutions, gave autonomous AI agents access to tools like email and messaging platforms, then documented what went wrong over 14 days. Vulnerabilities ranged from prompt-injection attacks to unauthorized data exfiltration and what the researchers described as “self-destructive behaviors.” “Prompt injection, agent hijacking, and rogue AI behavior aren’t theoretical anymore,” said Jennifer Johnson, CrowdStrike’s CMO. CrowdStrike is positioning its Falcon AI Detection and Response platform as the answer to these threats. Pre-registration for the challenge is currently open. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy https://cryptobriefing.com/editorial-policy/ .