New AIBOM Framework Automates Security Advisories for Agentic AI — Tested on 10,000 Components

The rise of agentic AI—autonomous systems that plan and act—introduces new security challenges, as traditional vulnerability management relies on static software bills of materials (SBOM). A new framework from researchers at Oxford, Cisco, and others addresses this by integrating AIBOM (AI Bill of Materials) with execution-bound runtime telemetry. The protocol-driven approach binds component declarations to deterministic environment captures and structured runtime data, enabling automated generation of CSAF VEX (Common Security Advisory Framework Vulnerability Exploitability eXchange) advisories. These advisories are computed from three inputs: declared artefacts, observed activation conditions, and enforced execution policies.

The system cryptographically signs each advisory and validates it through deterministic replay, ensuring reproducibility and trust. In evaluation, the framework handled approximately 10,000 component entries across synthetic agentic AI workloads ranging from 50 to 5,000 components, incorporating datasets from OSV, GitHub Advisory, KEV, and EPSS. This approach allows security teams to automatically generate exploitability assessments tailored to the actual runtime behavior of AI agents, rather than relying on static vulnerability databases. As agentic AI systems become more autonomous, such automated advisory generation is critical for maintaining security at scale. The research was published in Frontiers in Artificial Intelligence.