The Authorship Gap: Defending AI-Assisted Work When the Copyright Office Asks Who Made It

Abstract

The US Copyright Office's 2025 guidance on AI authorship resolved one question and created three more. It confirmed that human intellectual conception remains the legal center of gravity for copyright in AI-assisted works. It did not define what quantum of human contribution suffices, how to prove it after the fact, or how courts will weigh evidence generated by the same tooling ecosystem that produced the work. This whitepaper proposes an evidence framework — the Conception Chain — that maps each element of a copyright claim to specific, tamper-evident artifacts an enterprise can generate continuously rather than reconstruct under litigation pressure.

1. The Regulatory Landscape

Three developments define the current position of any company shipping AI-assisted creative or software output:

• The Thaler line of cases. Courts have consistently held that copyright protects only human authorship. Works produced without sufficient human contribution are public domain upon creation — not merely unregistrable, but unownable.
• USCO 2025 AI Authorship Guidance. The Office now expects registrants to disclose AI-generated portions and to be prepared to demonstrate the human intellectual conception behind protected elements. Disclosure is a filing requirement; proof of conception is a litigation requirement. Most companies have systems for neither.
• The training-data counterclaim wave. As plaintiffs' bars build cases against generative model providers, downstream users face discovery requests about their own AI tooling: which models touched the work, what prompts were used, and what human edits followed. A company that cannot answer these questions has a problem before it even reaches its own copyright claim.

2. Why Forensic Reconstruction Fails

The default enterprise response to an authorship challenge is reconstruction: pull git history, interview creators, export DMS version histories, and assemble a narrative. In our experience with diligence engagements, this approach fails for four structural reasons.

2.1 The Timestamp Problem

Filesystem and VCS timestamps are mutable by the very people whose credibility is at issue. A defense built on self-attested timestamps invites exactly the challenge it should prevent. Evidence requires an external, tamper-evident time source — not a local clock.

2.2 The Selection Problem

Reconstruction is inherently selective: teams preserve what they believe matters. Adversaries exploit the gaps. Continuous capture eliminates selection entirely — every iteration exists, including the ones the team would not have chosen to save.

2.3 The Attribution Problem

In AI-assisted workflows, a single artifact may pass through a model, a prompter, an editor, and a reviewer. Reconstruction typically captures the final state, not the human decision points — which is precisely where conception lives. The law asks what a human conceived, not what a pipeline output.

2.4 The Scale Problem

Modern products ship thousands of creative and code artifacts per quarter. No legal team can reconstruct hundreds of works after a claim arrives. Evidence must be generated as a byproduct of normal work, at the speed of the work itself.

3. The Conception Chain Framework

The Conception Chain decomposes an authorship defense into four evidence layers, each generated automatically and cryptographically linked to the next:

Layer 1 — Origin Attestation

When a human begins work on an artifact — opens a canvas, writes the first commit, drafts the prompt that seeds a generation — the tooling records a signed origin event: who, when (via an external time source), on what starting material. This is the seed of the chain and the direct answer to “where did this come from?”

Layer 2 — Conception Events

Human decision points are captured as they happen: manual edits, rejected generations, prompt revisions, explicit approvals. Each event is hashed and signed with a key held by the creator's organization. The layer distinguishes human contribution from model output — the exact distinction USCO disclosure requirements demand.

Layer 3 — Contamination Screening

For code and trained-adjacent assets, each artifact is screened against open-source license corpora at creation time. Findings are recorded in the chain, so a later auditor can see not only what licenses are present but when they entered and whether they were remediated.

Layer 4 — Publication Seal

At release, the full chain is sealed: a merkle root over all prior events, signed and timestamped, producing a single verifiable evidence packet. The packet is self-contained — anyone with it can verify every link without access to the originating systems.

4. Filing Strategy Under the 2025 Guidance

• Disclosure from the chain. USCO AI-portion disclosures can be generated directly from Layer 2 records, ensuring the filing matches the evidence rather than a summary of memory.
• Work-for-hire alignment. Conception events carry identity and employment context, supporting work-for-hire and assignment arguments for works created by contractors using company AI tooling.
• International posture. The same chain supports UK IPO and EUIPO filings; the EU AI Act's transparency obligations for AI-generated content map cleanly onto Layer 2 and Layer 4 records.

5. Litigation Posture

When a claim arrives, the Conception Chain changes the shape of the proceeding in three ways:

• Summary judgment surface. A sealed, externally timestamped chain is difficult to reduce to a fact question. Defenses anchored in verifiable records are better positioned for dispositive motion than defenses anchored in witness recollection.
• Discovery control. The chain answers the standard AI-tooling discovery requests (models used, prompts, human edits) from a fixed record, reducing depositions of individual creators — and the credibility risks that carry.
• Damages narrative. For works where ownership is established, the chain documents commercial exploitation timeline from first conception, supporting full damages rather than a contested slice.

6. Implementation Roadmap

Enterprises we advise typically deploy in three phases over one quarter:

• Weeks 1–2 — Capture. Integrate origin and conception capture into existing creative and code tooling. No workflow changes required of creators; evidence is a byproduct.
• Weeks 3–4 — Screen. Enable contamination screening in CI for all repositories and in the asset pipeline for creative output. Baseline findings become the remediation backlog.
• Weeks 5–8 — Seal and File. Produce publication seals for the current release train, generate USCO disclosures for pending filings, and archive evidence packets with counsel of record.

7. Open Questions We Are Tracking

• The sufficiency threshold: how much human contribution will courts require for heavily AI-assisted works? The chain's value grows as this question sharpens — granular conception records are the only way to argue the margin.
• Evidentiary standards for tool-generated evidence: expect Daubert challenges to automated capture. Our position is that cryptographic signing plus external time sources place such evidence on firmer footing than human testimony about digital events.
• International divergence: UK, EU, and US positions are converging on human-conception tests but diverging on disclosure mechanics. Multi-jurisdiction filers should keep chains jurisdiction-tagged from day one.

8. About AIthentica

AIthentica builds the evidence infrastructure described in this whitepaper: the Creative Process Provenance Suite for continuous conception capture, and the AI Code License Contamination Scanner for license exposure screening. The platform is SOC 2 Type II audited and ISO 27001 certified, with customer-held signing keys as the default posture.