Defining verifiable model registry requirements for dynamic systems

Specify the canonical information that should be stored in registries tracking frequently updated, multi‑component AI systems and develop mechanisms to verify the accuracy and integrity of registry entries over time.

Background

Modern AI systems are dynamic, comprising multiple evolving components, which creates oversight challenges.

Model registries could support tracking and verification, but it is unclear what information is necessary and how it should be validated to provide reliable accountability.

References

However, it's not clear what information should be stored in such a registry, nor how the information could be verified.

— Open Problems in Technical AI Governance  (2407.14981 - Reuel et al., 2024) in Section 5.3.2 “Verification of Dynamic Systems”

Managing the life cycle of attestable-build proofs and exploiting finer-grained dependency graphs for updates remain future work.

— Bootstrapping Mutual Attestation with Kleene's Second Recursion Theorem  (2608.20671 - Imamura, 21 Aug 2026) in Section 6, “Conclusion”

Supply-chain security remains a major open challenge in LLM systems. The complexity of modern LLM pipelines makes it difficult to verify the trustworthiness of all external components, particularly large-scale datasets and third-party model checkpoints.

— Shifting from Injection to Interaction: Rethinking Web Security in the Age of LLMs and Beyond  (2609.03999 - Singh et al., 3 Sep 2026) in Section 5, subsection LLM_03: Supply Chain Vulnerabilities, Open Research Challenges box