This issue explains how the two events are connected.

The month of August in time

Sony Music Publishing and Warner Chappell sued Anthropic on Friday, 28th August, in the Northern District of California, naming chief executive Dario Amodei and co-founder Benjamin Mann as individuals. The complaint covers tens of thousands of compositions, seeks up to $150,000 per willfully infringed work, and calls the conduct one of the largest ongoing thefts of intellectual property in history.

It joins the Universal, Concord, and ABKCO suit from January, which covers more than 20,000 songs and seeks more than $3 billion, the BMG case from March, and Round Hill's filing on 17th August.

Anthropic is preparing for a stock market listing in October, at a valuation investors expect to exceed $2 trillion. The new complaint sets its author payment claim against that number. It calls the $1.5 billion that Anthropic has already paid to publishers and authors a routine cost, not a meaningful check on the company's practices, because it amounts to a fraction of one percent of the valuation Anthropic is about to command.
However!
One demand buried in the complaint outweighs the rest.
The publishers want an account of Claude's training data, and no AI company can produce one. What entered a training corpus, from which source, and in what form, is being reconstructed years later through discovery, contested line by line, at a cost measured in billions and in delayed listings.

Every case on this docket turns on the same question!
What existed, in what form, before a machine used it.

The argument exists because nobody kept that record at the time.

ProofProfile kept it.
That is what we are selling.

The sale

The platform is for sale in full, as a single transaction, and full sales pitches begin in September.

The asset

The registry holds 2,006,172 notarized artists and their 9,900,355 recordings, each carrying a CopyrightChains registration and a separate Bitcoin timestamp.
A Bitcoin timestamp does one thing: it proves that a record existed in a specific form before a specific block on a chain no company controls. Anyone with an internet connection can check it independently.

Artist anchors sit in blocks 953,407 to 954,784. The closure manifest root entered block 961,779 on 9th August, 2026, and the whole set verifies together rather than record by record.

The source data underneath it is MusicBrainz and AcousticBrainz as they stood before 2022, the year generative music started being good enough to matter.
Every dispute now heading to trial concerns the era after that line.
The registry describes the era before it, sealed before Round Hill or Sony filed, so it cannot be accused of having been assembled for the litigation it now serves.

The trust

History in the registry is append-only. A record can be superseded by a newer one, but the supersession stays visible, and nothing is quietly overwritten. Cryptographic commitments for the entire set are published, so any third party can detect an attempted rewrite without first asking permission.

A provenance record that its owner can revise is a press release with a hash attached. That makes the question of who owns the industry's only independent record of the pre-generative era anything but academic. Custody decides who the record serves.

Inherited errors

The corpus carries the mistakes of its sources: wrong names, wrong credits, duplicates, all permanently sealed in. We cannot fix them, not for an artist who writes in, not for a buyer who asks, not even for ourselves. The moment we can correct one entry, we can change a date, and every date across ten million records stops being a fact and becomes an assertion.

The evidentiary boundary

The record establishes that a track existed in a specific form before a specific block. It says nothing about who wrote the song, who owns it, when it was made, or whether it was made by a human.
That refusal is structural, not a limitation.

A registry certifying human origin is one discovered AI-assisted entry away from worthless, because a single bad record poisons every claim beside it. A registry that certifies only dates cannot be caught out, because it never claimed anything that could turn out to be false.

The human judgment- who wrote it, who owns it, whether a person made it- belongs to a listener, a licensor, or a judge. They make that judgment better with a dated, unaltered record in front of them.

The publishers suing Anthropic will spend years and millions of dollars establishing facts of exactly this shape, one work at a time.

The answer layer

In May, a Munich court held Google directly liable for what its AI Overview said about two publishers, treating the summary as Google's own speech.

Google is appealing, and the injunction binds no one else, but every general counsel shipping AI answers has read it.

Two months later, on 31st July, the same court ruled against Suno in the GEMA case, the first European finding that unlicensed training on a society's repertoire infringes.

Two days after that, on 2nd August, the EU AI Act's transparency obligations took effect. A system that can be sued for its answers becomes careful about what it names, and names only what it can defend.

The advertising money followed the same path.

Google's search advertising grew 17 percent last quarter while its third-party network shrank.
On 24th August, ChatGPT began selling advertising across thirty-one European markets.
Advertising did not die with the link; it moved onto the surface producing the answer, and that surface now carries the liability.

Scarcity

Anyone can write the software; a competent team could ship an equivalent by Christmas, possibly a better one. Nobody can ship an older one. The attestations sit in blocks mined in June, against source data fixed before 2022, and on 9th August the supply of pre-seal attestations became permanently fixed.

Two majors have already bought the adjacent half of this problem. Warner acquired Sureel in June, which describes how models use a work's components. Universal's Music IP Holdings unveiled its patent framework on 20th August, licensing it first to Udio and GRAI, covering watermarking and licensed distribution. Both describe what happens to a recording during and after an AI event; neither establishes what existed before it, and the 28^th August 2026 hundred-page complaint is an account of how expensive that gap has become.

Contact

The pitch list includes Anthropic, Alphabet, OpenAI, Spotify, Thomson Reuters, Adobe, Apple, Amazon, Microsoft, Meta, RELX, Publicis, and Criteo, with two payment networks treated as conversations rather than approaches.
The majors are out of this process because those talks are already underway elsewhere.

Anthropic remains on the list for the same reason it dominates this issue: no company on earth currently has a stronger reason to own a sealed record of what existed before its models did.

If you know someone in corporate development, product leadership, or general counsel at any of them, we would like to hear from you.
If a name on that list is wrong, or an obvious one is missing, tell us which and why.

Contact: [email protected]
(Sorry, due to heavy spam and sales pitches, I only take calls from my contact list; send me an email first, and we'll set up a call)