New Open Source Tool Analyzes AI Models for Supply Chain Risks, Partners with MITRE and Indiana University
A new tool has emerged in the field of cybersecurity that aims to analyze artificial intelligence (AI) models for potential risk factors in the supply chain. Developed by the AI risk experts at Robust Intelligence, the tool, called the AI Risk Database, has been enhanced with new features and is now available as an open-source tool on GitHub. This move comes as part of a collaboration with MITRE and Indiana University, which will work together to enhance the database’s capabilities and integrate it with automated AI assessment tools.
Dubbed the VirusTotal for AI, the AI Risk Database is designed to assist the cybersecurity community in identifying and reporting security vulnerabilities found in public machine learning (ML) models. Additionally, it tracks other facets of these models that may jeopardize the reliability and resilience of AI systems, including issues related to brittleness, ethical problems, and AI bias.
The development of this tool stems from the growing concern surrounding supply chain risks in the world of AI systems. Similar to other aspects of software supply chains, AI systems rely on a wide range of open-source components to operate. However, the added complexity lies in dependencies on open-source ML models and data sets used for training purposes.
According to Hyrum Anderson, a distinguished ML engineer at Robust Intelligence and co-creator of the AI Risk Database, the reuse of models has greatly accelerated collaborative innovation. Nevertheless, it also means that a flaw in a single model can have far-reaching repercussions across numerous AI systems. This puts the spotlight on the need for AI supply chain security, not only for code and data but for models as well.
To advance the capabilities of the AI Risk Database, the tool will incorporate a new dependency graph feature developed by researchers from the Indiana University Kelley School of Business Data Science and Artificial Intelligence Lab (DSAIL). This feature will enable the scanning of GitHub repositories used to create models, identifying publicly reported flaws upstream of the delivered model artifact.
In addition to Indiana University’s involvement, a partnership with MITRE will further bolster the vulnerability research, classification, and risk scoring that powers the AI Risk Database. This integration will closely align the database with the MITRE ATLAS framework, an initiative led by MITRE to identify threats and risks to AI. ATLAS provides a comprehensive knowledge base that includes a list of adversary tactics and techniques based on real-world attack observations and AI red teaming.
Douglas Robbins, MITRE Vice President of Engineering and Prototyping, highlighted the significance of this collaboration, stating that the AI Risk Database enables organizations to understand their vulnerabilities when deploying specific types of AI-enabled systems. By hosting the database under the open-source MITRE ATLAS framework, this capability will play a crucial role in informing risk assessment and mitigation strategies for organizations worldwide.
To showcase the capabilities of the enhanced AI Risk Database, Robust Intelligence, MITRE, and Indiana University will hold a demonstration at Black Hat Arsenal. Hyrum Anderson, Christina Liaghati, the lead for MITRE ATLAS, and Sagar Samtani, director of Kelley’s DSAIL at Indiana University, will exhibit what the database can accomplish during sessions at Black Hat USA.
The development and release of this open-source tool represent a significant step forward in securing AI supply chains. By proactively identifying and addressing vulnerabilities in AI models, organizations can ensure the reliability and trustworthiness of their AI systems. With the collective efforts of Robust Intelligence, MITRE, and Indiana University, the AI Risk Database has the potential to become a crucial component in the cybersecurity toolsets used to combat AI supply chain risks.
