Building a Robust AI Bill of Materials for Enhanced Security

AI-BOMs: A Practical Guide to AI Bills of Materials

An AI Bill of Materials (AI-BOM) is a comprehensive inventory of an organization’s AI ecosystem, encompassing AI models, datasets, services, infrastructure, and third-party dependencies, along with their interrelationships. AI-BOMs utilize structured formats like SPDX extensions to facilitate sharing, auditing, and understanding of AI components across teams, similar to a software bill of materials (SBOM).

Differences Between AI-BOM and SBOM

While both AI-BOMs and SBOMs serve similar functions, AI-BOMs address the unique complexities of modern AI systems. Unlike the static nature of SBOMs, AI systems involve non-deterministic models, evolving algorithms, and data dependencies, which necessitate capturing these intricacies for effective AI Security Operations. An AI-BOM extends beyond mere code to include models, datasets, and dynamic dependencies.

The Necessity of AI-BOMs

The convergence of several factors has made AI-BOMs essential for responsible AI governance:

• AI risk and transparency demands: Organizations need visibility into the AI assets they employ to understand potential vulnerabilities.
• Regulatory pressure: New policies require meticulous documentation of AI components and their risk profiles.
• Supply chain security concerns: AI systems face risks from third-party models and APIs, necessitating thorough tracking.
• Internal governance requirements: Responsible AI initiatives require mechanisms for tracking model lineage and enforcing usage policies.

Real-World Example

In April 2024, researchers at Wiz uncovered critical vulnerabilities in Hugging Face’s AI-as-a-Service platform, which could have led to unauthorized access to sensitive data. A comprehensive AI-BOM could have identified these gaps, demonstrating the importance of maintaining visibility and continuous monitoring in AI systems.

Core Components of an AI-BOM

An effective AI-BOM comprises the following seven core components:

1. Data Layer: Captures all data assets essential for training and inference.
2. Model Layer: Tracks AI models, their metadata, and evolution over time.
3. Dependency Layer: Identifies vulnerabilities within the AI supply chain.
4. Infrastructure Layer: Monitors the hardware and cloud resources supporting AI workloads.
5. Security and Governance: Enables assessment of exposure and implementation of least-privilege access.
6. People and Processes: Documents ownership and change history across the AI lifecycle.
7. Usage and Documentation: Provides context on model behavior and performance metrics.

AI-BOMs and Security Functions

AI-BOMs facilitate key security functions such as:

• Discovery and inventory: Identifying all components within an AI environment.
• Traceability and explainability: Understanding model development and deployment.
• Risk assessment and prioritization: Evaluating exposure based on access and permissions.
• Governance and compliance: Supporting audits and regulatory requirements.
• Change management and incident response: Assessing impacts of updates and expediting investigations.

Compliance Frameworks

AI-BOMs serve as a foundation for meeting emerging AI governance requirements. They interact with frameworks like the NIST AI Risk Management Framework and the EU Artificial Intelligence Act, ensuring organizations can adapt to new compliance regulations efficiently.

Building an AI-BOM with Wiz

Wiz streamlines the development of AI-BOMs through:

• Automated discovery: Keeping AI-BOMs up to date as new services are deployed.
• Graph-based visibility: Mapping all AI components and their relationships.
• Policy enforcement: Integrating compliance checks into development pipelines.
• Drift detection: Monitoring changes to ensure compliance and security.
• Integration with workflows: Connecting with CI/CD pipelines for actionable insights.

Wiz positions AI-BOMs as a security-first system of record, enabling organizations to manage AI risks effectively while maintaining visibility and compliance. Request a demo to see how Wiz can enhance AI security operations from code to cloud.