Skip to main content
LEARN

AI Consent & Attribution

Machine-readable consent and proper attribution - the foundation for trust in the agentic web.

7 min read

Key Takeaways

  • Consent must be machine-readable for agents to respect it automatically
  • Attribution is recorded in receipts, creating verifiable credit chains
  • AIPREF (aipref.json) lets sites declare AI interaction preferences
  • Proper consent + attribution protects both content owners and AI operators

The Problem

The web was built for humans browsing with web browsers. Terms of service are written in legal English. robots.txt was designed for search engine crawlers, not AI agents that consume and transform content.

This creates two problems:

  • Content owners can't express preferences that agents understand - “training: no, RAG: yes, summary: yes with attribution” isn't something robots.txt supports.
  • Agents can't prove compliance - even well-intentioned AI systems have no way to demonstrate they respected consent or provided proper attribution.

Attribution in Practice

Attribution answers: “Where did this come from?” In agentic systems, this is captured through:

Source Recording

Receipts record the exact resource URL, timestamp, and content hash of accessed material

Credit Chains

When Agent B uses output from Agent A, the receipt chain traces back to original sources

License Compliance

Attribution requirements from AIPREF are embedded in receipts as verifiable commitments

Payment Proof

When attribution includes compensation, payment evidence is cryptographically linked

Standards & Protocols

The consent and attribution ecosystem includes several complementary standards:

  • AIPREF - AI preferences standard for declaring training, RAG, and usage permissions
  • PEAC Protocol - Policy discovery and verifiable receipts for agent interactions
  • C2PA - Content provenance standard for media authenticity and attribution
  • robots.txt - Legacy crawler control - still useful but insufficient for AI agents

Implementation

Get started with consent and attribution using Originary's tools:

Related Articles