Mitre ATT&CK Framework to Expand with AI Threat Techniques
Mitre's ATT&CK framework will incorporate new techniques for detecting and combating AI-driven cyberattacks in its upcoming release, aiming to catalog real-world adversarial behavior involving artificial intelligence.
Mitre, the federally funded nonprofit known for its ATT&CK framework, is set to enhance its widely adopted knowledge base with new techniques specifically targeting artificial intelligence (AI) driven cyberattacks. The upcoming Version 21, slated for release this spring, will integrate dozens of new techniques derived from actual incidents, providing cybersecurity professionals with more robust threat intelligence and detection strategies.
Adam Pennington, lead for the ATT&CK framework, announced the expansion at a recent conference, emphasizing that the core structure of the framework, which categorizes adversary tactics, will remain unchanged. He explained that AI agents, often trained by humans using established techniques, frequently employ the same methods as human adversaries. Consequently, defenders often struggle to differentiate between human-led and AI-generated attacks, as both can appear as sophisticated scripting or automated processes.
The framework's approach to incorporating AI threats will involve drawing from Mitre's other initiative, ATLAS, which focuses on AI-specific techniques. However, Pennington stressed that only techniques with documented real-world examples will be ported into the ATT&CK framework, maintaining its grounding in observed adversarial behavior. This ensures that the updates are practical and directly applicable to current threat landscapes.
While Version 21 will see a significant influx of AI-related techniques, the framework has already begun incorporating them. Version 20, released earlier this month, includes new techniques and sub-techniques such as prompt injection and AI-enabled data exfiltration. The planned enhancements for Version 21 are more ambitious, with AI poised to become a recognized platform within the framework's Enterprise Matrix, joining existing categories like Windows and containers.
Discussions are ongoing regarding the precise definition and scope of the AI platform. Lauren Lusty, who oversees the enterprise matrix, noted that open questions include whether the platform should be limited to AI models and training data, or extended to encompass the broader connections AI systems use within enterprise tools. Techniques that describe already well-known behaviors, such as gathering victim identity information, are unlikely to be candidates for inclusion as distinct AI platform techniques.
Mitre is actively soliciting feedback on the proposed AI platform, including through public meetings of the recently formed ATT&CK Advisory Council. This collaborative approach aims to ensure the framework remains comprehensive and relevant to the evolving threat landscape. Despite the focus on AI, Mitre maintains that the framework's development process remains largely human-driven, with manual analysis playing a crucial role in intelligence gathering and technique validation.
Pennington also shared insights into the quality of submissions received from researchers, noting that many submissions written by large language models lack the depth and analytical rigor required for inclusion. He highlighted that translating language patterns into effective intelligence analysis is a complex task that current AI tools still struggle to master, underscoring the continued importance of human expertise in cybersecurity.
This expansion of the ATT&CK framework reflects the growing sophistication and prevalence of AI in cyber warfare. By cataloging AI-specific attack vectors, Mitre aims to equip defenders with the knowledge and tools necessary to identify, understand, and mitigate these emerging threats, ensuring that cybersecurity strategies keep pace with technological advancements.