CrowdStrike is making Falcon AI Detection and Response (AIDR) generally available. With this expansion, the security company is targeting a relatively new attack surface: the interaction layer of AI systems. This includes prompts, AI agents, and the corresponding responses.
According to CrowdStrike, part of the security threat is shifting from traditional infrastructure to the way organizations apply generative AI.
Where existing security solutions focus primarily on data, models, and infrastructure, Falcon AIDR emphasizes the interaction between users, applications, and AI models. This interaction layer is becoming increasingly important as generative AI is used more widely within organizations, both during software development and by employees in their daily work.
Attacks on this layer include prompt injection, in which hidden instructions are added to input to cause AI systems to exhibit undesirable behavior. This can lead to the circumvention of security measures, the manipulation of output, or the exposure of sensitive information.
Same architectural approach as with EDR
According to analysis by SiliconANGLE, CrowdStrike applies the same architectural principles to AIDR as it did previously to endpoint detection and response (EDR). Instead of introducing separate security measures, AI security is integrated into a single platform that continuously collects and correlates telemetry. CrowdStrike thus positions the AI interaction layer as a fully-fledged part of the broader security domain, comparable to endpoints, cloud workloads, and identities.
Falcon AIDR is designed to give organizations insight into how AI is used within the enterprise. The platform records interactions with AI systems and agents, including through runtime logs that can be used for compliance and incident investigation.
In addition, the solution offers options for blocking AI interactions in real time when they are deemed risky. This includes detecting and stopping prompt injection attacks, jailbreak attempts, and other forms of unwanted or unsafe AI output.
An important consideration when using generative AI is the risk of sharing confidential information. Falcon AIDR can recognize sensitive data such as login credentials or regulated data and prevent it from being forwarded to AI models or external AI services. In doing so, CrowdStrike addresses concerns about data breaches and compliance when using AI within organizations.
Broader AI strategy
Falcon AIDR is part of the existing Falcon platform. In conversations with SiliconANGLE, CrowdStrike previously indicated that this development is in line with the broader Enterprise Graph strategy, which correlates large amounts of security telemetry to detect and interpret threats more quickly in the AI era.