Elastic has integrated the MITRE Adversarial Threat Landscape for Artificial-Intelligence Systems (ATLAS) framework into its security platform with the release of version 9.6. This update aims to provide security operations centers with specific visibility into threats targeting AI-enabled systems, augmenting the existing support for the MITRE ATT&CK framework. By incorporating ATLAS, the platform allows users to map prebuilt detection rules directly to AI-specific adversary tactics and techniques.
The integration focuses on common attack vectors identified by MITRE, including prompt injection, jailbreaking attempts, the theft of model API keys, and the poisoning of agent tools. Elastic confirmed that these ATLAS mappings are now applied to prebuilt detection rules designed for AI platforms and generative AI tooling, ensuring that security teams receive relevant contextual alerts without needing manual configuration.
To manage this data, Elastic has introduced a new managed MITRE data service within Kibana. This service facilitates the visualization of security coverage through a dedicated table that tracks how well existing rules address ATLAS tactics. Crucially, the platform separates these metrics from standard MITRE ATT&CK statistics, allowing organizations to assess their AI-specific security posture independently rather than having those figures diluted by broader enterprise threat data.
The ATLAS framework itself is a knowledge base maintained by MITRE, drawing on industry research and real-world observations of attacks against machine learning models, training data, and AI agents. While MITRE has announced plans to incorporate AI-enabled systems as a new platform within the broader ATT&CK Enterprise framework, ATLAS remains the primary, specialized repository for AI threat intelligence. Elastic’s implementation mirrors the structure of ATT&CK, utilizing tactics to define adversary goals and techniques to detail the specific methods employed.
Beyond detection, the update extends ATLAS support into the rule creation process and the Elastic Agent Builder. This allows developers and security engineers to incorporate AI threat context during the development of custom detection logic and the construction of context-aware agents. The company stated that this functionality is intended to help organizations identify which AI-specific threats their current security rules already mitigate and where gaps in coverage may exist.
This development arrives as the cybersecurity industry continues to formalize defenses against the unique risks posed by large language models and autonomous agents. By aligning its tooling with the ATLAS knowledge base, Elastic aims to standardize how security teams monitor the AI threat horizon, providing a structured approach to identifying and responding to malicious activity that targets AI infrastructure rather than traditional software environments.




