A distributor has begun removing independent artists' recordings on the strength of a major label's unproven pleading, telling the affected artists neither which recording allegedly infringed nor which rightsholder filed the claim. In the same fortnight that distributor was asked by a federal court to disclose its monthly volume of AI generated uploads in a case it is not a party to. Deezer will strip AI tagged music out of search results by the end of the year on a classification produced by its own detector rather than declared by whoever delivered the record, while Symphonic's own policy now says disclosure will not rest on what an uploader says. On 11 November the fee for filing the one kind of document that cannot be filed electronically, the old hand executed assignment, rises from $125 to $350. Three of the largest publishers in the world are meanwhile asking a federal judge to tell them which of their own songs were inside up to two million books that were scanned and then thrown away, which is the shape of the whole issue: decisions about somebody's rights are being taken in processes that person is not in.
Label Operations
A Major Label's Unproven Claim Is Now Enough To Take Your Release Down. Nobody Will Tell You Which Recording.
DistroKid has confirmed that it removed recordings from streaming services in response to claims made by Universal Music Group. Its own count is the phrase "a very small number of recordings". Amanda Ferri, the company's vice president of artist services, said DistroKid "strongly dispute[s] the allegations" in the Universal suit and "regret[s] the disruption to affected artists", adding that it was "working to minimize that disruption while protecting artists' ability to release their music". The only actual figure anywhere comes from an affected artist rather than from the distributor. The musician McGwire says six of his recordings came down. Universal filed against DistroKid in September and nothing in that case has been adjudicated, so every removal made so far rests on a pleading rather than on a finding.
The trade press led on the confirmation. The part nobody has pressed is the notice. On the artists' own account, many of the takedown notices named only "Universal Music Group" as the complaining rightsholder and identified no specific recording as the work allegedly infringed. That is a takedown in substance without the particulars a takedown is supposed to carry, which leaves the person on the receiving end with nothing to answer and nobody to answer it to. Jesse Cannon, a producer and music marketer, has been publicly critical of how DistroKid's support has handled the fallout while artists report strikes arriving with no warning beforehand and no reply afterwards.
One detail does more damage to the idea that any of this is about rights than anything in the pleadings. One of McGwire's recordings was taken down over an uncleared sample of a song that was released two years after his own track. A record cannot sample something that does not yet exist. Whatever produced that match, it was not a reading of the rights position. That it survived to the point of removal tells you how much human review sits between a major's claim and an independent's release coming off the shelf.
A claim nobody has had to prove, against a recording nobody will name, filed by a claimant identified only as a corporate group. There is no version of that an artist can answer.
Now look at the same distributor from the other direction. On 14 October Universal and Sony asked a federal court in Manhattan to compel DistroKid to disclose how many AI generated songs pass through its service each month, saying the figures would help establish market harm and accusing the company of refusing to produce documents sought under a subpoena served in September. DistroKid is not a party to that case. The defendant is Suno, the AI music company the labels have been suing since 2024. So inside a fortnight the same intermediary has pulled catalogue off the shelves on one major's unproven claim and been asked to account for its own delivery telemetry in a different major's case against somebody else. In neither proceeding is the label whose recording came down, or whose uploads are being counted, so much as a participant.
Two things follow for an independent and neither is expensive. Ask your distributor, in writing, what its review, appeal and reinstatement procedure actually is, because the reporting so far records no published route back from a removal of this kind. Then assume your delivery data is discoverable by parties you have no contract with, because it demonstrably now is. The aggregator chosen for convenience is the single point at which a catalogue can be removed, counted and produced to a stranger, which makes it the one relationship in the chain that most independents have never papered beyond accepting the terms on the signup page.
Sources: Hypebot, 12 October 2026, for the Ferri statement, the McGwire removals and the sample dated after the track · Law Commentary, on the removals · Law Commentary, 16 October 2026, for the 14 October motion to compel and DistroKid's non party position in the Suno case · Digital Music News, 16 October 2026
Metadata & Standards
Three Platforms Now Decide Whether Your Record Is AI. Only One Of Them Starts By Asking You.
Deezer will add a switch before the end of the year that takes fully AI generated tracks, albums and related artists out of search results altogether. Chief executive Alexis Lanternier framed it as letting listeners "remove this content completely from their streaming experience on Deezer". The volumes behind it are not marginal: Deezer said in July that more than half of everything delivered to it each day was AI generated, averaging around 90,000 tracks, against roughly a tenth of daily uploads when it switched its detection tooling on in 2025. The thing to notice is where the classification comes from. Deezer built that detection tooling itself, so the tag deciding whether a record appears when somebody types the artist's name is one the service produces rather than one the deliverer declares, with no appeal route for a misclassified human recording appearing in the announcement or in any of the coverage of it.
Symphonic comes at it from the contract instead. Its generative AI policy requires clients to disclose AI generated and AI assisted material and may require them to name the tools, models or companies used. It lists false AI disclosure among the things it will reject or restrict, which makes an inaccurate declaration a breach in its own right rather than a subset of inaccurate metadata. Two provisions matter more than the obligation itself. The first is that disclosure can be combined with technical review, fingerprinting, provenance and detection systems, meaning a label can be contradicted by a detector on a release it declared accurately. The second is that acceptance by Symphonic "does not guarantee acceptance by every DSP, territory, program or monetization opportunity", so a release can clear the distributor and still be excluded at a platform, leaving its owner holding a delivery that is live in some stores and absent from others for reasons sitting in two separate policies.
YouTube Music's delivery route sits between the two. A record arrives carrying one of four states: fully generative AI, partly generative AI, no generative AI, or no supplied value at all. The fourth is the one worth thinking about, because nothing published for labels says what happens to an empty field and an empty field is still a value. What counts as "partly" is not defined in anything published for labels either, which means the single most consequential word in the declaration is the one nobody has specified.
Every one of these platforms now has an AI policy. Not one of them gives a label a field it controls, a definition it can apply and a published route back from a wrong answer.
For a label the practical ask is narrow. Decide the position on every release before it is delivered, record the reasoning somewhere that survives staff turnover, then keep the session evidence, because under all three regimes the burden of showing that a record is human made has already moved to the person who made it. The field that declaration goes in, meanwhile, arrived with no announcement at all, which is the first item on the wire.
Sources: TechCrunch, 14 October 2026, for the Deezer toggle, the Lanternier quote and the upload volumes · Reading of Symphonic's generative AI policy, 23 September 2026, for the disclosure obligation, the verification language and the DSP acceptance clause · Reading of YouTube's music gen AI declaration routes, updated 14 October 2026
Policy & Regulation
The Only Copyright Filing You Cannot Make Online Is About To Cost $350. It Is Also The One Diligence Always Turns Up.
The United States Copyright Office submitted its new fee schedule to Congress on 14 July 2026. Congress has 120 days in which to disapprove it, failing which the Register may institute it, which puts the operative date at 11 November. The Office has said it aims to implement in the autumn. None of this is secret and the trade coverage has been accurate as far as it goes, which is the cost of registering a new release.
Recordation is the part that bites a catalogue transaction and it is the part that moves furthest. Electronic recordation of a document goes from $95 to $215. The base fee for everything that has to be filed on paper goes from $125 to $350. A paper notice of termination goes from $125 to $275. Set those three beside each other and the schedule is pricing format rather than complexity, because the documents that cannot be filed electronically are precisely the old ones: the hand executed assignment from 1979, the deed with a wet signature, the chain of title that exists as photocopies in a box in somebody's loft. Those are exactly the documents a buyer's diligence surfaces as well as the ones a seller discovers it never recorded.
There is a second thing in the schedule that nobody has written about and it splits the two halves of the same release. Group registration of an album of sound recordings doubles, $65 to $130. Group registration of the musical works on an album goes to $85, which is not only lower than the recording side but lower than the $130 the Office itself had proposed for it at the notice of proposed rulemaking stage. The Office moved in both directions during the review. From 11 November, registering the recordings on a record will cost roughly half again what registering the songs on it does, which is a cost signal pointing straight at the publishing side being the cheaper thing to keep current.
A fee schedule is not usually an editorial subject. This one puts a date and a price on the consequence of never having filed the paperwork.
The window is the point rather than the percentages. Anything sitting unrecorded in a drawer is cheaper to file in the next three weeks than in the fourth, where the filings most likely to be sitting there are the paper ones whose fee is nearly tripling. A catalogue that has been meaning to record its older assignments has been handed a deadline it did not choose and roughly three weeks in which to use it.
Sources, read at source on 18 October 2026: US Copyright Office, Proposed Schedule and Analysis of Copyright Fees, for every figure and the 14 July submission date · Hypebot, 13 October 2026, for the 11 November date and the end of the congressional review period · Federal Register, 91 FR 13529, 20 March 2026, the notice of proposed rulemaking
The Cautionary Tale
Three Publishers Are Asking A Court To Tell Them Which Of Their Own Songs Were In The Books
Concord, Universal Music Publishing Group and ABKCO sued Anthropic in January 2026 in the Northern District of California over more than 20,000 songs. On 11 September they asked the court to let them investigate something called Project Panama inside the current discovery phase rather than a later one. Project Panama is an internal Anthropic programme that surfaced when planning documents were unsealed in the separate authors' case in late July. On the reported account of those documents, the plan was to buy up to two million used books from the retailers Better World Books and World of Books, slice the bindings off with a hydraulic cutter, scan the loose pages with industrial scanners, then send the paper to a recycling plant.
Among the material acquired were songbooks and sheet music collections. The physical copies were discarded after scanning. No list of what was in them has been produced by anybody, which is why the motion asks a judge to compel one. Three of the largest music publishers in the world are in the position of being reasonably confident that their catalogue was taken while being unable to enumerate what was taken, because the object that identified each work was a printed book and the book no longer exists.
The version of this that happens in an independent's life is smaller and the mechanism is identical. A work is only as assertable as the record identifying it, so when the identifying artefact is the physical thing itself, a shelf of masters, a box of lyric sheets, a hard drive, a songbook, losing the artefact does not merely lose the copy. It loses the ability to say what was owned. A catalogue described only by its contents cannot be described at all once the contents are gone. Note the provenance honestly: the Project Panama detail and the destruction method come from reported accounts of the unsealed documents rather than from the motion, which has not been published.
Sources: Read Dork, 17 September 2026, for the discovery application and the destruction of the physical copies · Schrodinger's Mood, 17 September 2026, for the Project Panama planning detail, the suppliers and the 11 September filing date, carrying a pseudonymous byline over a named footer credit
On The Wire
Four shorter things. Two of them nobody reported.
Metadata & StandardsThe field every platform AI tag depends on arrived in a point release with no date on it
DDEX's release notes for the Electronic Release Notification suite list seven versions and carry no publication date against any of them. Generative AI signalling, meaning the ability for a sender to communicate that a recording was made fully or partly by a generative AI tool and that a contribution was made by one, was introduced in version 4.3.1. It is described in two lines of an informative annex, alongside a classical title field and the renaming of Genre to DisplayGenre. The current top version, 4.3.2, does exactly one thing: it corrects the datatype of a brand identifier element. So the capability underpinning every platform AI declaration regime in this issue entered the delivery standard as a minor revision nobody announced. The narrow checkable question for a label is which ERN version its own delivery chain actually emits, because anything below 4.3.1 has nowhere to put the declaration at all.
TechnologyHalf the music tech companies surveyed say most of their development goes on working round broken data
Eleven of twenty two companies surveyed by the Association For Electronic Music named conflicting metadata across databases as their single biggest structural problem. A further 41 per cent named the absence of universal artist and song identifiers as a core one. Half said that more than half of their total development resource goes on workarounds rather than on core product, while the report puts the number of collection societies that rights are spread across at more than 120. Chloe Dagenais, quoted in it, reduces the whole thing to two sentences: "We've got data standards. But no one is enforcing those standards." Treat the figures as what respondents said rather than as measured, since the sample is 22 self selected companies with no published questionnaire or response rate. It remains the clearest available answer to a label owner asking why the tooling in this sector is so thin, which is that half the sector's engineering capacity is being spent on reconciliation instead of on building anything.
Policy & RegulationThe Supreme Court considered the worldwide termination question on Friday and the answer lands the day before this issue
BMG Rights Management (US) LLC v Vetter, No. 25-1391, was distributed on 30 September for the conference of 16 October, which was last Friday. The docket shows no order as this is written. An order list from that conference is therefore expected on Monday 19 October, the day before this issue publishes. The question is whether a termination under US law reclaims worldwide rights or only US rights, which the Fifth Circuit answered worldwide. The two possible outcomes point in opposite directions: a denial leaves that ruling standing with no national resolution, while a grant sets it up to be reversed and puts every worldwide licence written against a terminated grant back into question. Either way the exposure for a rights holder sits in the same place, which is whether its own works schedule records the territory each grant actually conveyed, rather than assuming the answer is everywhere.
Read AcrossFrom 2 December an AI generated recording is supposed to arrive carrying its own provenance
The EU's Code of Practice on Transparency of AI generated content requires providers to mark outputs using at least two machine readable techniques, typically digitally signed tamper evident metadata together with imperceptible watermarking. Audio is expressly in scope under Article 50(2), which covers synthetic audio, image, video or text. Generative systems already on the market have until 2 December 2026 under the transition the AI Omnibus inserted, with watermark detection interoperability following on 2 February 2027. Around 190 organisations had signed by the end of July. Two cautions: the Code is voluntary and sits alongside the binding obligation rather than replacing it, while no specific technical standard is named anywhere, so there is still no published answer to which marking counts. The operational read is narrow enough to act on. From December a file arriving with no provenance metadata at all is itself a signal, while the prohibition on stripping that metadata falls on the model providers rather than on everybody downstream of them, so the open question for a label is whether its own delivery chain preserves the marking or quietly removes it. Nobody has published a position on that, which makes asking the only way to find out.
And Finally
Facebook Told The Man Who Wrote The Song That Rights Holders Had Detected Their Music In It
From February this year rather than this week, because it has not stopped being the neatest illustration of the problem. Jireh Lim wrote "Buko" in 2013 and released it himself, uploading it to YouTube where it went viral. In February he posted a Reel using his own recording and Facebook flagged it. The notice told him that multiple rights owners had requested changes to the video because a large amount of their music had been detected in it and that they would be claiming the Reel. It then reassured him that the claim applied to that one Reel, leaving his Page and his own monetisation untouched. He replied, in Filipino, that the person who actually wrote that song was him. Eighteen thousand reactions agreed with him.
The joke is entirely in the facts, which is how this slot prefers it. The point underneath is exact. A fingerprint match routes a claim to whoever is registered against the recording and the author has no standing anywhere in that flow, because to the claim system an author is simply another uploader who happens to match. If the person who wrote the song cannot assert authorship inside the claim process, then the process is not testing authorship at all. It is testing registration.
The second beat is older and even better on ownership recitals. In January 2024 the Trademark Trial and Appeal Board decided who owned the name of a band called Split Decision. Billy Stott had managed the group since 1991, a year after two other people formed it, then claimed the name on the strength of a management agreement saying the band's names were owned solely by the Manager, plus a title he had awarded himself: "primary, non performing member". The agreement named no band. The Board held that an agreement saying you own something does not make it so and decided ownership instead on who the public associated with the mark and who it looked to for the quality of the performance. Stott lost on all three factors. The same sentence does the work in catalogue diligence: an ownership recital is not a chain of title, while a schedule that does not identify the asset conveys nothing.
Sources: Interaksyon, Philstar, 25 February 2026 · ArentFox Schiff, 8 March 2024, on Billy Stott v Split Decision Music LLC, Opposition No. 91249613, decided 31 January 2024 and non precedential
Feature Spotlight
Works And Publishing: The Register Is The Only Version Of Your Rights Anybody Else Can See
Read the issue back and the same absence does all of the damage. A recording comes down because a claim was routed against a registration rather than against a right. A distributor is asked to count uploads and the label whose catalogue it is never hears about it. Three platforms classify a record as AI or human and none of them gives the deliverer a field it controls. A fee schedule puts a price and a date on assignments that were never recorded. Three publishers cannot enumerate their own works. A songwriter has no standing in a claim against his own song. In every one of them the operative record belongs to somebody else and the rights holder's own version of it either does not exist or cannot be produced quickly enough to matter.
Works and Publishing is the part of rightsHUB that holds the works themselves: writers and their splits, publishers and their shares, the identifiers that let a work be matched, the registration position society by society and the territory each grant actually conveyed. It exists so that the answer to who owns this and on what terms is a record you maintain rather than a reconstruction you attempt under pressure. It produces a clean CWR export music publishers can file rather than a spreadsheet somebody has to rekey, which matters because an unregistered work is, to every system described above, an unclaimed one.
The question worth asking is the one every story in this issue answers badly. If a platform classified one of your releases tomorrow, or a major named it in a claim, or a buyer asked for the works schedule, how long would it take you to produce the document that settles it. The rest of the toolkit is at rightshub.net/features.
You do not get to choose which record is authoritative. You only get to choose whether yours agrees with it.