Introduction of AI Bill of Materials (AIBOM) for Enhanced AI System Security

The AI Bill of Materials (AIBOM) has been proposed as a standardized framework to enhance the transparency and security of AI systems. This framework extends the existing Software Bill of Materials (SBOM) to include specific AI-related artifacts such as model lineage and training provenance. A proof-of-concept platform, AIBoMGen, automates the generation of signed AIBOMs, ensuring integrity through cryptographic methods. Evaluations show that AIBoMGen can detect unauthorized modifications and generate AIBOMs with minimal performance impact. The implementation of AIBOM is crucial for compliance with regulatory frameworks like the EU's AI Act, which mandates strict documentation for AI systems. The lack of transparency in AI models poses risks, especially in sensitive environments, making the AIBOM framework essential for establishing trust and accountability. The proposed methodologies aim to operationalize AIBOM within Trusted Research Environments (TREs), addressing gaps in software provenance and vulnerability exposure. Overall, AIBOM represents a significant step towards securing AI systems and ensuring their responsible use.

Key Points: • AIBOM is designed to enhance transparency and security in AI systems. • AIBoMGen automates the creation of signed AIBOMs, ensuring artifact integrity. • Compliance with frameworks like the EU's AI Act is facilitated by AIBOM's structured documentation.