Data usage evidence

What did the agent do with the work?

Licenses and AI preference standards say how a work may be used. These records show what an agent actually did with it, signed so the creator can check. Every access, every refusal and every output links back to the work and the person behind it.

What gets recorded

Run it

The noncommercial access demo shows the whole loop on a synthetic work marked CC BY-NC 4.0. The same agent asks twice, with two declared purposes. The commercial request is refused before retrieval. The noncommercial one is allowed, and its output carries attribution back to the work. A standalone verifier checks every signature and link, and seven tampered bundles each fail.

git clone https://github.com/aeoess/noncommercial-access-demo
cd noncommercial-access-demo
npm install
npm start
npm run verify

Where it fits

Other efforts already name the work and state the terms. The IETF AI Preferences work defines how preferences for AI uses are expressed. RSL states licensing and compensation terms. Creative Commons licenses and CC signals set the conditions creators choose. ISCC gives a work a content-derived identifier that records can bind to. These records add the evidence of what happened, built to sit next to all of them.

The closest related proposal is draft-reilly-aipref-compliance, which defines compliance records for retrievals, aggregated under a signed Merkle root so a publisher can audit only its own records. The records here add the agent's side of a request, which that draft does not address. They carry the delegation behind the request, a refusal made before retrieval, and attribution of what the agent produced.

What it does not do

Code