By Crystal Morin, Senior Cybersecurity Strategist, Sysdig
AI investment isn’t slowing down. According to Accenture’s Pulse of Change report, 86% of C-suite leaders plan to increase AI spending in 2026, with most considering it a revenue driver rather than an expense. KPMG also found that 65% of UK respondents plan to continue investing in AI regardless of measurable ROI. AI has become a long-term strategic priority for many organizations.
Security, however, has not kept pace. Research by Spacelift has found that while many organizations express confidence in their ability to govern AI, far fewer have formal governance processes in place. The gap between confidence and implementation is where many risks remain.
Effective AI governance begins with understanding what makes up an organization’s AI environment. Establishing an AI Bill of Materials (AIBOM) is an important first step.
What goes into an AIBOM
An AIBOM builds on the foundation of the more familiar Software Bill of Materials (SBOM), extending its scope to cover the full AI stack. It includes core elements such as the large language model (LLM), as well as the surrounding components that influence outputs, including prompt-processing systems, developer tooling, model weights, training data, and external data sources.
When multiple AI agents work together and each model has its own AIBOM, organizations must also account for the connections between them. These interactions are becoming increasingly standardized through the Model Context Protocol (MCP) and MCP servers.
Creating an AIBOM is only part of the process. Keeping it current is often more challenging. AI systems evolve rapidly as developers release updated models and APIs to improve performance. An AIBOM should reflect those changes and maintain a record of what was updated and when.
This can be difficult in fast-moving development environments where speed of deployment often takes precedence over governance. Yet accurate inventories become even more important during periods of rapid change.
Why AIBOMs matter for governance and security
Every time an AI application is used, it interacts with multiple components across the AI stack before generating a response. In agentic systems, information may pass through several agents before reaching the end user. If something goes wrong, or if an audit is required, organizations need visibility into those interactions.
An AIBOM provides that baseline by documenting which components were involved in processing a request and what data sources contributed to the response. If an AI system produces inaccurate or sensitive information, those records can help investigators understand how the output was generated. They may also support legal or regulatory reviews by documenting how data was handled.
Organizations cannot effectively manage risks associated with components they have not identified. Each element of an AI stack introduces its own security considerations, and attackers have increasingly targeted AI infrastructure. For example, researchers observed an authentication vulnerability in the open-source multi-agent orchestration framework PraisonAI being exploited within hours of its public disclosure. Similarly, a server-side request forgery (SSRF) vulnerability in LMDeploy, a toolkit for serving vision-language and text-based large language models, was reportedly exploited shortly after disclosure. Although neither project was among the largest AI platforms, both became targets quickly.
As the time between vulnerability disclosure and exploitation continues to shrink, maintaining an up-to-date inventory helps organizations identify affected components and prioritize remediation. Runtime security measures can further help organizations detect and respond to active attacks while patches are being deployed.
Planning ahead
Today, AIBOMs remain an emerging best practice rather than a widely adopted or mandatory requirement. Even so, they can help organizations maintain accurate inventories and strengthen AI governance.
Guidance around AIBOMs is evolving. Earlier this year, G7 countries released guidance on Software Bills of Materials for AI, while the Five Eyes countries published best practices focused on Secure by Design and Secure by Default principles for AI infrastructure.
As AI adoption continues to expand, AIBOMs are likely to play a growing role in AI governance and security. Their effectiveness, however, depends on organizations maintaining accurate inventories and incorporating them into their broader governance practices.
Emerging guidance from governments and industry groups is likely to encourage broader adoption, but organizations that already have a clear understanding of the components in their AI environments will be better positioned to manage security, compliance, and operational risks.