Copyright law is clarifying its position on AI-generated content faster than most practitioners expected. The US Copyright Office has issued guidance. Federal courts have ruled. The EU AI Act has added disclosure obligations. The legal landscape is settling, but the practical question for most organizations is not who the law says owns AI-generated content. It is whether they can prove it.
Where Copyright Law Currently Stands
The US Copyright Office's position has been consistent since its February 2023 guidance and subsequent policy updates: copyright protection requires human authorship. AI-generated content produced without creative control by a human is not eligible for copyright protection. Content that uses AI tools but involves sufficient human creative expression can be copyrightable, covering the human-authored elements.
The courts have followed this analysis. In Thaler v. Vidal and related proceedings, courts confirmed that non-human entities cannot hold intellectual property rights. In Thaler v. Perlmutter, the district court upheld the Copyright Office's refusal to register a work created autonomously by an AI system, ruling that human authorship is a constitutional requirement for copyright.
This creates a two-tier landscape. Content businesses that use AI tools under meaningful human direction, selecting, arranging, and modifying AI outputs with creative judgment, can claim copyright in the resulting work. Organizations that deploy AI to generate content autonomously, without substantial human involvement, may find that content is not protectable.
The Getty Images Litigation and What It Signals
Getty Images filed suit against Stability AI in the UK and the US, alleging that the training of Stable Diffusion on Getty's licensed image library constitutes copyright infringement. The case is ongoing, but the allegations and the legal theory being tested matter to any organization that trains or fine-tunes AI models on third-party content.
The outcome will affect how organizations approach model training data, licensing, and the downstream use of outputs from models trained on potentially infringing data. Until there is settled law, organizations using outputs from models trained on unlicensed third-party content carry unquantified exposure.
Deepfakes and Publisher Liability
The liability landscape for AI-generated synthetic media is developing rapidly. The US has enacted the DEFIANCE Act, creating civil liability for non-consensual intimate images generated by AI. Multiple states have enacted their own deepfake legislation covering electoral contexts, synthetic media in advertising, and other specific use cases. The UK and EU are developing analogous frameworks.
For publishers and media organizations, the liability question is not hypothetical. Publishing AI-generated synthetic media of real persons, without consent and without disclosure, creates exposure under an expanding body of law. The EU AI Act's Article 50 imposes mandatory disclosure requirements for AI-generated content, including synthetic media, intended to inform the public about matters of public interest.
Why Watermarking Is Insufficient
Watermarking is the most commonly proposed technical solution to the attribution problem. It has real limitations. Digital watermarks can be stripped by post-processing. Resizing, compression, format conversion, or deliberate adversarial attacks can remove or degrade watermark signals. Research has demonstrated that current AI watermarks are not robust against determined removal.
Watermarks are also not timestamped. A watermark identifies an output as AI-generated but does not record when it was created, what model produced it, or who authorized it. For evidentiary purposes, the timestamp and the chain of custody are often more important than the origin flag.
And watermarks are not independently verifiable. Verifying a watermark requires access to the watermarking key or system maintained by the operator. If the operator disputes the record, or is no longer available, independent verification is not possible.
What On-Chain Provenance Provides
On-chain attestation creates a timestamped, tamper-evident, independently verifiable record of an AI output's creation. The hash of an output committed to a blockchain at the time of creation provides a proof of existence at a specific block timestamp. Any party can verify the record without relying on the operator to maintain it.
For copyright purposes, this provides the evidentiary foundation that watermarking cannot. It does not create copyright protection, but it creates the record that makes ownership claims auditable and defensible. The difference between "our AI system produced this content on this date" and "here is an independently verifiable on-chain attestation of that fact" is the difference between a claim and evidence.
Mintlayer's IP Notary provides on-chain certificate sealing for AI outputs, creating provenance records that can be produced in legal proceedings, regulatory examinations, and licensing disputes.
This article is for informational purposes only and does not constitute investment advice.
Mintlayer Web Services helps organizations build on Bitcoin-native infrastructure. Learn more →
