A city wants AI-generated open-meeting summaries marked so public copies can be traced if altered. Which technique directly supports that authenticity goal?

Marking AI-generated open-meeting summaries so public copies can be traced if altered is watermarking for civic content authenticity tracing. Watermarks or related integrity marks help agencies detect tampering and attribute machine-generated public records more reliably than unmarked text alone, which supports open-government trust when summaries circulate beyond the clerk’s office. Hashing model weights to support civic content authenticity tracing fails by concept because hashing model weights fingerprints the model binary for version verification; it does not place any traceable mark on the public copy of the summary itself, so an altered public copy still can't be authenticated against it. Signing audit logs to support civic content authenticity tracing fails because signed logs protect the integrity of internal audit records, not authenticity marking of the published artifact. Attaching disclaimer headers to support civic content authenticity tracing fails because disclaimer headers only add a removable notice that content is AI-generated; they still provide no content authenticity signal if a published copy is later altered or republished. Exam caveat: watermarks vary in robustness; determined adversaries may strip weak marks, so pair watermarking with signed distribution channels and authoritative publication locations. Operational check: apply an approved watermarking or signing method at generation time, publish summaries from an official portal, retain unmarked masters under access control, and test detection on deliberately altered public copies.