The Elvis lyrics suit: the credit-stripping claim is the long shot worth backing

The headline number in the Presley suit against Anthropic is $48.6M. The part worth watching is the $6.85M of it that rests on one quieter idea: that stripping a songwriter's name from a lyric before feeding it to a model is its own wrong, separate from copying the lyric. We think that claim is a long shot. We also think it is the right claim to bring.

The news hook

On 23 October, ABG EPE IP LLC, part of Authentic Brands Group, joined by Raleigh Music Publishing, Abilene Music and Aberbach and Creatore family interests, sued Anthropic in the Northern District of California over lyrics to 274 songs Presley recorded, Music Business Worldwide reports. Alongside copyright infringement and circumvention, the complaint says song titles, songwriter credits and publisher details were removed before training. According to MBW, it alleges Anthropic tested text-extraction tools and picked Newspaper over jusText, which kept footnotes with copyright notices and owner names that Anthropic called "boilerplate" and "useless junk". Anthropic told MBW it disagrees with the claims and will defend the case fully.

In our news post we said the credit claim mattered more than the damages total. This column tests that call against the strongest objection.

The best case against us

The law here is section 1202 of the US Copyright Act. It bars anyone from intentionally removing or altering "copyright management information" without authority, and that information includes a work's title, its author's name and its copyright owner (17 U.S.C. § 1202). On paper, a lyric with its writers' names cut off fits neatly.

In practice, AI defendants have mostly beaten these claims, and the reasons line up uncomfortably with this complaint. A Skadden review of the 2024 decisions sets them out:

• No harm, no case. In Raw Story v. OpenAI, Judge Colleen McMahon dismissed for lack of standing, holding that "the DMCA merely protects the integrity of a work's CMI, not how it is used," and that the publishers showed no concrete injury.
• No identical copy, no claim. In Andersen v. Stability AI, the court held a removal claim "requires identicality between the original work and the copy". The same question went to the Ninth Circuit in the GitHub Copilot case.
• The one survivor had output. The Intercept v. OpenAI got its 1202(b)(1) claim past Judge Jed Rakoff, but its amended complaint showed ChatGPT continuing its articles.

Now set the Presley complaint beside that. MBW notes it contains no example of Claude reproducing any of the 274 songs; it alleges infringing output "on information and belief", and its lead rewrite example comes from the Round Hill Music suit and doesn't involve a Presley song. On the opposing view, this is Raw Story with a better-known catalogue: a credit claim with no shown copy and no shown harm, bolted onto a template complaint to lift the damages total. Backing it, the argument goes, is backing the weakest part of the case.

That is a fair argument, and on the pleadings as reported it may well win.

Our answer

We still think the claim is worth bringing, for three reasons.

It goes at intent, which is what the statute asks for. Section 1202(b) turns on intentionally removing the information. Most AI-training complaints have to infer that from the bare fact of scraping. This one, per MBW, points to a choice between extraction tools, with the tool that kept copyright notices rejected and those notices called junk. If the discovery record bears that out, it is closer to direct evidence of the mental element than earlier plaintiffs had. It doesn't fix the identicality problem, but it makes it harder to wave the claim away as an accident of engineering.

The harm is easier to name for songwriters than for news sites. McMahon's point was that the publishers in Raw Story couldn't show what removing their bylines cost them. For songwriters, credits aren't decoration; they are how royalties, splits and future work get routed. A lyric that circulates without its writers' names is a lyric that is harder to get paid for. That is our inference, not something the courts have accepted, but it is a more concrete injury than the one Raw Story rejected.

It is the claim that survives a fair-use loss. The training-is-fair-use fight is being had in bigger cases, like the Concord publishers' case against Anthropic, and this suit will likely follow whatever those decide. A credit claim asks a different question: even if training is lawful, may you deliberately cut the names off first? A clear answer to that would reach every company that scrapes lyrics, which is why it matters well beyond 274 songs.

What we think

Expect the credit count to be the first thing Anthropic attacks at the motion-to-dismiss stage, and don't be surprised if it goes, at least until the plaintiffs can show output. But the publishers are right to keep bringing it. Copyright damages are a fight over money already lost; the credit claim is a fight over whether songwriters stay attached to their words as models absorb them. For where this suit sits among the rest, see our AI music lawsuits explainer.