Versioning is operational for public model repositories such as Hugging Face Hub, and release/lifecycle documentation is increasingly institutionalized through EU AI Act duties such as high-risk-system technical documentation and post-market monitoring, GPAI model technical documentation, and Commission lists for GPAI models with systemic risk. Separately, at least one AI lab publishes release-specific system cards or safety reports in the evidence reviewed here. But the paper’s harder target—verified dynamic-system identity across model weights, prompts, tools, retrieval corpora, routing, safety filters, and frequent updates—is not solved by the sources reviewed here. They provide documentation, histories, lists, and release reports, but they do not establish a general cryptographic proof that a hosted API is serving the same model, prompt stack, tools, retrieval corpus, routing, and filters that auditors evaluated.