Skip to main content

Proof of Creativity Protocol (PCP)

How do we attribute value when AI creates?

Proof of Creativity addresses the IP attribution problem for AI-generated content — tracking provenance, attributing contributions, and flowing value to creators in the generation chain.

The Problem

When AI generates content:

  • Who owns it?
  • What training data contributed?
  • How does value flow to original creators?
  • How do we prevent plagiarism at scale?

Story Protocol

story.foundation — Building the IP infrastructure for AI.

Three components form the stack:

ComponentFunctionWhat It Proves
IP RegistrationOn-chain record of creative workThis existed before AI touched it
LicensingProgrammable terms for derivative useThis agent had permission to use it
Royalty DistributionAutomatic value flow to contributorsThe original creator got paid

Each component is a smart contract. Registration creates an immutable provenance record. Licensing encodes terms that agents can read programmatically. Royalties flow automatically when derivatives generate revenue.

Commerce Flow Position

PCP is the EVALUATE stage in the agent commerce flow:

DISCOVER → COMMUNICATE → COMMERCE → AUTHORIZE → SETTLE → EVALUATE

PCP

After settlement, PCP answers: was the output original? Did value flow to contributors? Does the creator's reputation change? Without this stage, agent commerce has no quality signal feeding back into discovery.

The Extraction Test

When an agent generates content, three questions determine whether PCP applies:

  1. Derived from identifiable source? — If yes, provenance chain required
  2. Commercially distributed? — If yes, licensing terms must be satisfied
  3. Revenue generated? — If yes, royalty distribution triggers

If all three are no, PCP is unnecessary overhead. If any are yes, skipping PCP creates legal and ethical liability that compounds with scale.

Context

Questions

When AI-generated content has no identifiable source — trained on billions of fragments — does the concept of "original creator" still apply?

  • If licensing terms are programmable, who writes the terms for training data that predates the protocol?
  • At what scale does manual IP registration become impossible, and what replaces it?
  • Does PCP create a moat for early registrants, or does it level the field for all creators?