As Bloomberg first reported, Apple has chosen to use Google’s AI model to power the new Siri. Apple will pay Google $1 billion a year for use of its AI model.
Meanwhile, Google pays Apple $20 billion a year to be the default search engine in Safari.
Why this deal makes sense
Putting aside the performance of the various models, another reason this deal makes sense from a business standpoint is that Google has the least exposure to the Shadow Library copyright claims compared to OpenAI and Anthropic, the other companies that Apple reportedly also consider for Siri.
Both OpenAI and Anthropic used pirated copies of books from so-called shadow libraries. (Indeed, the same person, Benjamin Mann, was reportedly responsible for that undertaking at both OpenAI and Anthropic.) In Bartz v. Anthropic, that use of shadow libraries was deemed not fair use by Judge Alsup, resulting eventually in a $1.5 billion settlement. A similar fate may await OpenAI.
Google doesn’t face the same Shadow Library allegation based on use of Books3, Library Genesis, PiLiMi, or some of the other datasets allegedly used in the litigation against OpenAI and Anthropic. Instead, the 2 lawsuits against Google includes an allegation that Google copied (through crawling) from websites (Z-Library, PDFDrive, and OceanofPDF): “Google relied on automated systems to make unauthorized copies of the Plaintiff Works and millions of others, creating common questions of law and fact that predominate over any individual issues. Google did not selectively target or differently handle specific Plaintiff Works— rather, it applied consistent methods across Plaintiff Works that it copied to build and train its Generative AI Models.”
There is no allegation of use of a “shadow library,” such as Library Genesis.
Presumably, that’s because Google has its own internal library of books from its Google Book project based on scanning of lawfully purchased copies of books held at libraries. In Bartz, Judge Alsup said Anthropic’s similar scanning and digitizing copies of books was a fair use. In two prior lawsuits, the Second Circuit held that book search projects (by Google and Hathitrust) were fair use.
This doesn’t mean that Apple is clear and free of copyright litigation. Book authors have filed 3 lawsuits against Apple for its alleged use of Shadow Library copies of books as indicated in a research paper by Apple researchers.
But, if Apple’s choice was among the models of Anthropic, OpenAI, and Google, one attractive feature of Google’s model is that it has the least exposure to the Shadow Library infringement claim.
Although it’s possible the newer models of Anthropic and OpenAI didn’t rely on Shadow Libraries for their training, why risk having to prove that in copyright litigation if a third option forecloses that copyright claim altogether.
But, if Apple’s choice was among the models of Anthropic, OpenAI, and Google, one attractive feature of Google’s model is that it has the least exposure to the Shadow Library infringement claim.
Related stories