AI Provenance Failures Link Derbyshire Evidence, Medical Upcoding, and Model Export Controls

Three unrelated beats expose the identical structural flaw: generative systems now produce outputs that enter high-stakes pipelines without verifiable origin or audit. The Derbyshire constable fabricated case evidence with AI and submitted it as real; PwC’s 2026 report shows AI coding tools systematically inflating billed severity for unchanged care; Commerce Department export curbs on Anthropic’s frontier models were triggered after Jassy meetings precisely because the models’ weights and training data could not be reliably tracked once released. In each instance the same missing layer—cryptographic provenance and immutable logging—allowed unverified AI artifacts to cross into legal records, insurer ledgers, and adversary hands. The pattern is not sector-specific risk but a single supply-chain defect now manifesting wherever AI output is treated as authoritative input.