A new arXiv paper proposes anchoring AI security agents to a blockchain to keep the software supply chain honest.
Academic researchers describe a system that coordinates several specialized AI agents, each backed by a large language model, to watch different parts of the software development lifecycle: source integrity, dependency and SBOM analysis, CI configuration auditing, artifact verification, and runtime policy checks. Every agent generates a cryptographically signed attestation, which gets recorded on a permissioned blockchain through smart contracts covering an agent registry, an immutable attestation log, and a release-policy module. A consortium-operated certificate authority secures communication between agents and blockchain nodes. The paper's use case walks through a source-code security agent running its analysis, anchoring the result on-chain, and triggering an automated allow-or-block deployment decision.
Supply chain attacks keep succeeding because the tools meant to catch them - SBOMs, CI scanners, artifact checks - don't always agree with each other or leave a trail anyone can verify after the fact. This framework tries to fix that by giving every automated check, including the AI making the call, a signed and immutable receipt. It also answers a newer worry: as agentic AI takes over more security monitoring, something has to audit the auditors.
For now, this is a design paper with a sequence diagram, not a deployed system - the real test is whether it can handle the sprawl of dependency graphs on npm or PyPI, not a single walkthrough use case.