CrowdStrike, in collaboration with Amazon Web Services (AWS), has announced "AI Unlocked: Agents of Chaos," a $100,000 international AI security challenge. This virtual red teaming game is designed to help organizations understand and secure the emerging enterprise attack surface presented by AI agents. By simulating real-world threats like prompt injection and agent hijacking, the initiative provides defenders with hands-on experience in managing autonomous machine identities. For B2B leaders, this signals a critical shift toward securing AI agents that execute code and access enterprise systems at machine speed.
The Agents of Chaos Red Teaming Competition
The $100,000 competition is structured into three distinct Acts, offering progressive challenges and prize pools. Act 1, "The Sanctum," runs from August 31 to September 7 with a $10,000 prize. Act 2, "The Gatekeeper," follows from August 31 to September 14 with $20,000 at stake. The final and largest phase, Act 3, "The Basilisk," takes place from September 15 to September 29, offering a $70,000 prize. Players compete by using prompt injection and other techniques to manipulate weaponized AI agents within a fictional environment. Participants earn points for solving puzzles, though scores are adjusted based on the number of tokens consumed by their prompts. While the competition is accessible virtually worldwide, an on-site gameplay experience is also planned for Fal.Con 2026. This gamified approach aims to teach defenders how to stop rogue AI behavior and unauthorized actions before they impact actual enterprise environments.
Securing Autonomous Machine Identities and AI Agents
As enterprises accelerate AI deployment, every autonomous agent introduces a new identity to govern and a new attack surface to defend. CrowdStrike notes that traditional security frameworks were not designed to protect autonomous machine identities that possess real credentials and move data at machine speed. These agents can escape controlled environments and breach systems they were never intended to access. The "Agents of Chaos" challenge specifically targets these vulnerabilities, such as agent hijacking and prompt injection, which allow adversaries to trick models into revealing information or taking unauthorized actions. By focusing on runtime security, the initiative highlights the necessity of protecting AI agents as they execute code and interact with enterprise systems. This focus aligns with CrowdStrike’s positioning as a pioneer in AI Detection and Response (AIDR), emphasizing the need for specialized security measures to manage the risks inherent in the agentic era of computing.
Key Takeaways
- The $100,000 international challenge is divided into three Acts: The Sanctum ($10,000), The Gatekeeper ($20,000), and The Basilisk ($70,000).
- The competition utilizes red teaming techniques, including prompt injection, to simulate how AI agents can be manipulated to perform unauthorized actions.
- The virtual game will be live internationally beginning August 31, with an on-site experience scheduled for Fal.Con 2026.
TechInsyte's Take
In our view, this collaboration between CrowdStrike and AWS underscores a critical realization for the C-suite: AI agents are no longer theoretical risks but active operational liabilities. As these autonomous entities gain the ability to execute code and access sensitive data using real credentials, they bypass many traditional identity and access management protocols. This challenge signals that the industry is moving toward a "runtime-first" security mindset for AI. For decision-makers, the implication is clear: securing the AI lifecycle requires more than just perimeter defense; it requires specialized detection and response capabilities capable of monitoring machine-speed interactions and preventing agentic drift or hijacking.
Questions & Answers
How does the "Agents of Chaos" challenge simulate real-world enterprise risks?
The game uses red teaming scenarios where players employ prompt injection and other manipulation techniques to trick AI agents into revealing information or taking unauthorized actions, simulating how autonomous models can breach systems and escape controlled environments.
What are the specific timelines and prize structures for the competition?
The competition runs from August 31 to September 29 across three Acts: Act 1 (Aug 31–Sept 7) offers $10,000; Act 2 (Aug 31–Sept 14) offers $20,000; and Act 3 (Sept 15–29) offers $70,000.
Why is traditional security considered insufficient for the new AI agent attack surface?
Traditional security was not built to manage autonomous machine identities that operate at machine speed, execute code, and use real credentials to access enterprise systems and move data, which can lead to agents breaching unauthorized systems.
What is the strategic significance of the collaboration between CrowdStrike and AWS?
The partnership combines CrowdStrike's AI Detection and Response (AIDR) expertise with AWS's infrastructure to address the emerging security needs of organizations deploying autonomous AI agents within enterprise environments.
Source: BUSINESSWIRE
