CrowdStrike Launches AI Security Challenge to Test Defenses Against Agentic AI Threats

CrowdStrike Launches AI Security Challenge to Test Defenses Against Agentic AI Threats with the introduction of AI Unlocked: Agents of Chaos, a global cybersecurity competition developed in collaboration with Amazon Web Services (AWS). The initiative is designed to give security professionals hands-on experience defending against emerging AI threats such as prompt injection, agent hijacking and unauthorized AI behavior.

CrowdStrike has unveiled AI Unlocked: Agents of Chaos, a global AI security competition aimed at helping cybersecurity professionals understand and defend against emerging attacks targeting autonomous AI agents. Developed in collaboration with Amazon Web Services (AWS), the virtual challenge combines gamified learning with real-world offensive security techniques, offering participants a chance to compete for $100,000 in prizes.

The announcement highlights a growing concern across the cybersecurity industry as enterprises rapidly adopt AI agents capable of executing workflows, accessing enterprise systems and interacting with sensitive business data. While these autonomous systems improve productivity, they also introduce new attack surfaces that traditional security tools were not designed to protect.

Unlike conventional cybersecurity exercises focused on networks or endpoints, AI Unlocked: Agents of Chaos places participants in a fictional adversarial environment where they must exploit AI agents using techniques such as prompt injection, information disclosure and agent manipulation. The objective is to demonstrate how attackers can influence AI behavior while teaching defenders how to recognize and mitigate these emerging threats.

Prompt injection has quickly become one of the most discussed security risks associated with generative AI. By crafting malicious or deceptive prompts, attackers can manipulate AI models into ignoring instructions, revealing sensitive information or performing unintended actions. As organizations increasingly deploy AI agents capable of interacting with cloud infrastructure, enterprise applications and internal data, protecting these systems has become a growing priority for security teams.

The competition is structured around three progressive missions, or Acts, where participants infiltrate a fictional cybercriminal organization deploying weaponized AI agents. Players must manipulate those agents to uncover hidden information, bypass protections and prevent a simulated AI-driven attack before time runs out. Scoring is based on successful task completion while minimizing the computational resources, or tokens, consumed by prompts, encouraging efficient and effective exploitation techniques.

The challenge begins on August 31 with Act 1: The Sanctum, followed by Act 2: The Gatekeeper, before culminating in Act 3: The Basilisk, which carries the largest prize pool. Participants can compete remotely worldwide, with an additional in-person gameplay experience available during Fal.Con 2026, CrowdStrike’s annual cybersecurity conference.

Beyond the competitive element, the initiative serves as a practical demonstration of an emerging cybersecurity discipline known as AI Detection and Response (AIDR). CrowdStrike has positioned AIDR as an evolution of traditional detection and response technologies, focusing specifically on identifying, monitoring and securing AI agents operating within enterprise environments.

As AI becomes embedded across software development, customer service, business automation and cloud operations, every autonomous agent effectively becomes a new machine identity requiring governance, authentication and continuous monitoring. Unlike human users, AI agents can execute actions at machine speed, making compromised or manipulated agents capable of causing significant operational impact in a short period.

According to Gartner, organizations are expected to significantly increase investment in AI governance and security as generative AI becomes integrated into core business operations. At the same time, IDC projects continued growth in enterprise spending on AI-powered cybersecurity solutions designed to secure increasingly complex hybrid and cloud-native environments.

The cybersecurity industry has responded with growing investment in AI security platforms from major technology providers including Microsoft, Google Cloud, Amazon Web Services (AWS) and specialized cybersecurity vendors. These platforms increasingly focus on securing AI models, protecting machine identities, monitoring prompt interactions and detecting anomalous AI behavior before it leads to business disruption.

Red teaming has also emerged as a critical practice for evaluating AI systems. Rather than testing only network defenses or software vulnerabilities, AI red teaming examines how language models and autonomous agents respond to malicious inputs, adversarial prompts and unexpected operational scenarios. Exercises such as Agents of Chaos provide defenders with practical exposure to these techniques in a controlled environment.

As enterprises continue deploying AI-powered assistants and autonomous agents, cybersecurity professionals face the challenge of securing systems that can independently access enterprise resources, execute workflows and interact with sensitive information. Hands-on training initiatives are becoming increasingly valuable as organizations build the expertise needed to defend against this rapidly evolving threat landscape.

By combining education, competition and real-world attack simulations, CrowdStrike’s AI Unlocked initiative reflects the industry’s broader shift toward preparing security teams for the unique risks introduced by autonomous AI systems. As agentic AI adoption accelerates, understanding how these technologies can be manipulated—and how to defend them—will become an increasingly important component of enterprise cybersecurity strategies.

Top Insights

• CrowdStrike and AWS launched AI Unlocked: Agents of Chaos, a global AI security competition with a $100,000 prize pool focused on defending AI agents.
• Participants gain hands-on experience exploiting and securing AI agents through prompt injection, agent hijacking and AI red teaming scenarios.
• The challenge highlights the growing importance of AI Detection and Response (AIDR) as enterprises deploy autonomous AI across business operations.
• AI agents are creating new enterprise attack surfaces, increasing demand for runtime protection, AI governance and machine identity security.

Have a story or business inquiry? Contact our CyberTechEdition team today.