Mirror Protocol is a proposed architecture for capturing disclosure, attribution, and auditability in AI-assisted creation workflows. This page provides a staff-friendly mapping: legislative intent → system gaps → proposed function. It also states plainly where the bill's scope and this framework's scope differ.
AI systems can generate creative and derivative outputs at volumes that overwhelm legacy disclosure, registration, licensing, and enforcement processes. Without a creation-time governance layer, the market risks a new black box—this time in intellectual property provenance.
Status: specified and provisionally patented. Not yet built. No production deployment exists.
This is an implementation mapping of a proposed architecture. It is not legal advice, and nothing in the third column is a shipping product. Read the scope note below before the table.
The CLEAR Act regulates model developers at the input layer: what copyrighted works went into training. Mirror Protocol addresses creators at the output layer: what a person can demonstrate about work they made. These are different duties on different parties, and this framework does not satisfy the bill's requirements on a developer's behalf. What the two share is a structural assumption — that a record of origin exists — which neither the bill nor current practice specifies how to produce or verify. The table below maps that shared assumption, not equivalence.
| Legislative Intent | Current System Gap | Proposed Mirror Protocol Function |
|---|---|---|
| Training-data disclosure by developers | Disclosure is often vendor-private, inconsistent, or reconstructed after disputes. Records are not standardized. | Out of scope. This obligation falls on model developers and is not addressed by this framework. Listed here for completeness rather than alignment. |
| Machine-reviewable reporting | Most provenance data is unstructured text, scattered across tools, or missing completely. | Audit Trail Export (proposed): machine-readable logs on a consistent schema, producing repeatable evidence bundles. |
| Attribution clarity (human + AI contributions) | Multi-tool workflows lose chain-of-custody between platforms; “who did what” disappears. | Attribution Ledger (proposed): timestamps, contributor roles, and version history across tools and sessions. |
| Reduce litigation-by-forensics | Current practice often becomes post-hoc investigation rather than creation-time compliance. | Creation-Time Governance (proposed): compliance artifacts generated during creation rather than reconstructed later. |
| Support scalable registries and disclosure systems | Legacy systems were designed for human-speed volumes, not AI-scale generation. | Standardized Output Layer (proposed): consistent records capable of feeding databases, dashboards, and audits. |
Development status: the architecture is specified and covered by a provisional patent filing. It has not been implemented, and no working demonstration is available at this time. Nothing above should be read as a description of deployed software.
This page describes a technical architecture approach and does not interpret legal obligations or offer legal advice.
The argument for a record of origin is no longer speculative. In the largest copyright settlement in United States history, the entire question turned on where the material came from.
N.D. Cal., No. 3:24-cv-05417. A $1.5 billion class settlement covering roughly 500,000 works obtained from the LibGen and PiLiMi datasets and used in AI training. The court held that training on lawfully acquired books is fair use; the class was certified for piracy, not for the act of training.
The same training, on the same books, producing the same model, was lawful or unlawful depending entirely on how the copies were acquired. Provenance was not a side issue in that case. It was the case.
The honest limit: Bartz concerns training inputs, not creative outputs — a different layer from the one this framework addresses. It is cited for what it establishes about the category, not as precedent for this work.
This is offered as context, not as a claim about the CLEAR Act's scope. The bill addresses training-data transparency in AI model development. But the infrastructure required to satisfy it—creation-time provenance, portable attribution, machine-readable audit trails—is the same infrastructure a newsroom needs to prove a story's chain of custody. One build, two constituencies.
A 2026 UNESCO-commissioned evidence brief by Bunce and Pearson estimates the annual global cost of disinformation at US$355–516 billion, and reports a randomised trial of AFP fact-checking finding roughly an 8% reduction in the circulation of checked material. Both figures are estimates and are sourced in full on the evidence base. The distance between an eight per cent correction effect and a problem of that scale is the argument for governance at the creation layer rather than the moderation layer.
See the evidence base and sourcing →A defensible record from assignment to publication: intent logged before the first word, every tool that touches the story identified, provenance gaps flagged before release, and an exportable evidence bundle if the work is later disputed—with source identity severed from the exportable record. This describes the design, not a deployed system.
Open the journalism application →I’m requesting a short technical briefing with legislative staff to discuss implementation pathways for disclosure-ready provenance and audit trail systems, and where the creator-side layer sits relative to the bill's developer-side requirements.
These are documents. There is no working demonstration to show at this stage, and we will not represent otherwise.
AI is not merely faster. It changes the economics of creativity and the scale of knowledge reuse. When disclosure, attribution, and auditability are missing, the vacuum gets filled by opaque control—black boxes, fragmented systems, and permission-based learning. Mirror Protocol is designed to keep the system transparent, interoperable, and accountable as the next wave arrives.
Multi-AI workflows don’t naturally coordinate. Governance must travel with the work across tools, vendors, and versions.
Transparency is not just compliance—it’s the mechanism that protects creators and stabilizes markets under AI-scale output.
The durable solution is middleware: disclosure pipelines, attribution ledgers, verification services, and exportable audit trails.