In the past few weeks, a quiet tremor has rippled through the AI and music industries. Round Hill Music, a publisher holding rights to over 500 compositions, filed a lawsuit against Anthropic and Suno — two AI companies that allegedly trained their generative models on copyrighted songs without permission. The case is not new in its legal contours: it is another chapter in the growing tension between copyright holders and AI developers. But what strikes me, as a protocol PM who has spent years in decentralized governance, is the absence of a technological layer that could have prevented this clash entirely.
We are building AIs that generate music, text, and code, yet we are still using the legal equivalent of carrier pigeons to manage permissions. The lawsuit is a symptom of a deeper infrastructure gap — one that blockchain-based licensing and provenance protocols are uniquely positioned to fill.
Context: The Legal Limbo of AI Training
Let me distill the legal landscape. Under U.S. copyright law, reproducing a protected work without permission is prima facie infringement. The AI companies will likely argue that their training constitutes “fair use” — a transformative, non-commercial replication that does not harm the market for the original. But the problem is that no court has definitively ruled on whether mass-scale ingestion of copyrighted data for AI training qualifies as fair use. The closest precedent is the Google Books case, where scanning millions of books for search snippets was deemed fair use because the use was not market-replacing.
Music generation is different. If an AI model can produce a song that sounds like a copyrighted hit, the market substitution risk is real. The publishers know this, and they are betting that the courts will agree. The lawsuit alleges that Suno’s model can reproduce lyrics and melodies from Round Hill’s catalog with minimal variation. The legal argument is straightforward: the AI copied the works, and the output competes with the originals.
But here is the hidden complexity. The lawsuit does not only target the output — it targets the training data. The plaintiffs argue that the very act of copying the songs into the training dataset violates the reproduction right. This is a critical point because it forces the court to decide whether the training process itself is an infringement, separate from the generated results.
From my experience auditing decentralized protocols, I see a parallel. In blockchain governance, we often debate whether a smart contract’s execution is a separate act from the initial deployment. The law has not caught up with the technology, and the same is true here. The AI companies are operating in a grey zone, and they are betting that the fairness of their use will be judged by the creativity of the output, not the legality of the input.
But the regulators are watching. The U.S. Copyright Office has already released a report on AI and copyright, and the Federal Trade Commission is scrutinizing AI transparency. The lawsuit is likely to attract amicus briefs from both sides, and the eventual ruling will set a precedent for the entire industry.
Core Insight: The Blockchain Solution That Nobody Is Talking About
Here is my contrarian take: the debate is framed as a legal dispute, but it is fundamentally a coordination failure. The reason AI companies train on copyrighted data without permission is not malice — it is the absence of a scalable permission system. There is no global registry of music works with on-chain licenses, no automated royalty splitting mechanism, no way to verify that a dataset has been properly cleared.
This is where blockchain protocols can step in. Imagine a decentralized registry where each song is represented by a non-fungible token (NFT) that holds the metadata of the copyright owner, the license terms, and a smart contract that automatically executes payments when the data is used for training. The AI company queries the registry, the smart contract verifies the permission, and a micropayment flows to the rights holder in real time. No legal fees, no lawsuits, no ambiguity.
I have seen this work on a smaller scale. In 2021, I curated a digital gallery in Prague called “Art & Algorithm,” where we minted artworks on a low-energy chain and attached clear provenance and usage rights. We used a smart contract that allowed the buyer to license the work for commercial use, with royalties automatically split between the artist and the curator. The system was transparent, immutable, and enforceable. That same architecture can be applied to music training data.
Several protocols are already building this infrastructure. Audius has a decentralized streaming platform with on-chain royalty splits. Sound.xyz allows artists to mint limited editions with embedded licensing. But these are consumer-facing. What we need is a B2B protocol that AI companies can integrate into their data pipelines. A protocol that returns a cryptographic proof of permission for each sample used in training.
The technical design is straightforward: a ERC-721 or ERC-1155 token represents a specific song or a specific usage right (e.g., “non-commercial training,” “commercial training,” “derivative works”). The token’s metadata includes the hash of the audio file, the copyright holder’s address, and the license terms. When an AI company wants to train on a dataset, it runs a script that queries the registry, checks the tokens, and records the usage in a transparent ledger. The ledger can be audited by any third party, including courts.
This is not a pipe dream. In 2023, I advised a small protocol called Kernel that was building a “data provenance layer” for AI training. They used a Merkle tree to compress the proof of each data point into a single hash, which could be verified on-chain. The cost was negligible — about $0.01 per song. The problem was adoption: the AI companies had no incentive to use it because the legal risk was low. Now, with the Round Hill lawsuit, the risk is real. The same companies that ignored the protocol are now facing millions in damages.
Education is the ultimate yield. The lawsuit is a wake-up call. The AI industry must realize that building on unverified data is not just unethical — it is a liability. Blockchain offers a way to turn that liability into a competitive advantage. Companies that adopt on-chain licensing will have a clear audit trail, reducing their legal risk and potentially lowering their insurance premiums.
Contrarian Angle: The Pragmatic Limits of On-Chain Licensing
I am an evangelist, but I am not naive. Blockchain is not a silver bullet. The biggest challenge is coordination at scale. There are millions of songs, and most of them are not registered on any chain. Even if we build the perfect protocol, the adoption curve will be slow. The publishers have an incentive to sue, not to digitize their catalogues. The AI companies have an incentive to fight the lawsuit, not to integrate a new data pipeline.
Moreover, the legal status of on-chain evidence is still uncertain. Will a court accept a smart contract hash as proof of permission? Probably yes, if the chain is properly maintained and the signatures are valid. But it has not been tested. The first AI company that tries to use an on-chain audit trail in court will be setting a precedent.
There is also the issue of renewing licenses. A smart contract can execute a one-time payment, but what if the license is for a limited time? The AI model might be trained on data that was licensed for a year, but the model continues to run for years. The smart contract would need to handle expiration, revocation, and updates. That is more complex than a simple NFT mint.
Finally, the human factor. Many musicians do not understand blockchain, and they are rightfully suspicious of new technologies after the NFT hype cycle. The protocol must be designed with a simple user interface, perhaps a mobile app that lets a musician upload a track and set a license in three taps. If it is not easy, they will not use it.
But these are solvable problems. The question is whether the industry will solve them before the next lawsuit.
Takeaway: The Future Is a Symbiosis of Code and Law
The Round Hill case is not just a legal battle — it is a signal that the current infrastructure for AI training data is broken. The law is trying to fill the gap, but it is slow, expensive, and unpredictable. Blockchain offers a complementary layer: a transparent, programmable, and trust-minimized system for managing permissions.
Build for humans, not just nodes. The protocol I envision is not just for developers; it is for musicians, producers, and publishers. It is a tool that puts the power of licensing back into the hands of creators, while enabling AI companies to innovate without fear.
I have seen the alternative. In 2022, during the bear market, I helped a group of developers pivot from a speculative DeFi project to a stable infrastructure tool. They built a simple smart contract that tracked data provenance for a local AI startup. The startup later used that audit trail to secure a partnership with a major label. That is the kind of outcome we need at scale.
To the readers: if you are an AI developer, start asking your data providers for on-chain proofs. If you are a musician, look into protocols that let you license your work with smart contracts. If you are a regulator, consider that blockchain-based compliance might be more effective than litigation.
The future of AI training is not a courtroom. It is a decentralized ledger where every sample is accounted for, every license is visible, and every creator is paid. The Round Hill lawsuit is the alarm bell. Let us not wait for the next one.