Dear CIO,

The convergence of artificial intelligence (AI) and Software Bills of Materials (SBOMs) is redefining the cybersecurity playbook. As software systems grow more complex, the traditional SBOM, which catalogs components and dependencies, is evolving into something far more sophisticated: the AI Bill of Materials, or AI-BOM. These next-generation inventories are purpose-built to account for AI systems’ unique risks and components. In this newsletter, we will cover the intersection between AI and SBOMs.

Best Regards,
John, Your Enterprise AI Advisor

AI & SBOMs: The New Blueprint for Software Transparency

Unlike traditional SBOMs, AI-BOMs capture a broader spectrum of elements like datasets, models, algorithms, hardware environments, supporting libraries, and even ethical considerations. This comprehensive visibility is crucial for addressing AI-specific risks like model tampering, data poisoning, and vulnerabilities in obscure machine learning libraries. Simply put, AI-BOMs offer the level of transparency needed to protect today’s increasingly autonomous and opaque systems.

Just as AI presents new risks, it’s also becoming a powerful ally in managing them. AI-enhanced tools can now automate the creation and maintenance of SBOMs, providing more accurate, real-time insights than ever before. These tools can parse source code, binaries, and package manifests at speeds and depths that human reviewers simply can't match.

One standout example is ERS0, a platform that uses AI for deep binary analysis and string-matching techniques to identify firmware components, even those hidden behind obfuscation. Another example is AI-powered behavioral analysis tools that are adding context to risk assessments by interpreting how software functions in the real world, building on frameworks like MITRE ATT&CK.

This is where stacks like NORMAL come into play. Designed to bring structure to modern AI development, NORMAL emphasizes observability, rigorous model management, and advanced techniques like Retrieval-Augmented Generation (RAG). With observability platforms like Arize and Langsmith, NORMAL helps teams monitor AI behavior, detect anomalies, and ensure consistent performance. Its emphasis on version control and lifecycle governance makes it a natural fit for AI-BOMs, aligning engineering practices with security and compliance needs.

This being said, challenges remain. AI-BOMs are only as good as the data they contain, and maintaining accuracy, currency, and integrity is no small feat. What’s more is that in fast-moving AI environments, models and datasets change frequently, sometimes daily. Cryptographic validation and secure attestations are becoming essential tools for ensuring the authenticity and reliability of BOM data.

Regulatory pressure is also raising the stakes. In the U.S., Executive Order 14028 mandates machine-readable SBOMs for federal use, signaling a broader shift toward mandated transparency. The EU AI Act takes this a step further, requiring detailed documentation for high-risk AI systems. This effectively demands AI-BOMs in all but name.

Despite these pressures, many organizations are playing catch-up. Building and managing AI-BOMs requires expertise across AI, cybersecurity, compliance, and ethics, a rare combination that many companies are still working to assemble.

Looking forward, expect BOMs to become “living” documents that are updated dynamically as systems evolve. Standards like SPDX and CycloneDX are already expanding to include AI-relevant fields, signaling that AI-BOMs are becoming a foundational layer for software security and governance.

Bottom Line for CIOs

AI is changing everything, including how we track and secure software. The emergence of AI-BOMs reflects a broader shift toward accountability and transparency in machine learning systems. CIOs who lean in now to invest in tools like NORMAL and set up strong governance will be building trust, compliance readiness, and a competitive edge in an AI-driven future.

• Chris Hughes warns that the greatest AI-related security risk lies in security teams failing to proactively embrace, enable, and innovate with AI.

• Emma Woollacott looks at a study revealing that 42% of employees now secretly use AI daily to gain an edge.

• Todd Bishop highlights an AWS study showing generative AI's surge to the top of IT budgets.

• Ethan Steiniger argues that addressing AI hallucination starts with improving retrieval via better context.

• Rakesh Gohel compares emerging AI agent protocols by breaking down their architectures, use cases, and communication models.

• Bill Toulas demonstrates the real-world risk of CVE-2025-30065 in Apache Parquet.

• Ross Kelly writes on new research, which cautions that despite widespread optimism about generative AI, 39% of UK tech leaders view board expectations as unrealistic.

• Helen Oakley announces the first open-source tool for generating AI SBOMs on Hugging Face.

• The Artificially Intelligent Enterprise looks at how to stay relevant, productive, and 10x more valuable with AI.

• AI Tangle covers OpenAI backing down from its plans to become for-profit.