CTI4AI: Threat Intelligence Generation and Sharing after Red Teaming AI Models

One such early As the practicality of Artificial Intelligence (AI) and Machine Learning effort is MITRE ATLAS (Adversarial Threat Landscape for Artificial- (ML) based techniques grow, there is an ever increasing threat Intelligence Systems) knowledge base [8] modeled after the MITRE of adversarial attacks. There is a need to'red team' this ecosystem ATT&CK framework [9]. ATLAS includes a well-defined overview to identify system vulnerabilities, potential threats, characterize of adversary tactics, techniques, and case studies for AI systems properties that will enhance system robustness, and encourage the based on real-world observations and demonstrations from AI security creation of effective defenses. A secondary need is to share this groups, and from academic research. AI security threat intelligence between different stakeholders like, In this paper, to overcome the need to methodically identify and model developers, users, and AI/ML security professionals. In this share AI/ML specific vulnerabilities and threat intelligence we create paper, we create and describe a prototype system CTI4AI, to overcome and describe a prototype system CTI4AI. The system leverages the need to methodically identify and share AI/ML specific DARPA's GARD AI red teaming toolkit to identify vulnerabilities vulnerabilities and threat intelligence.