Google hit with $270M fine in France as authority finds news publishers’ data was used for Gemini
In a never-ending saga between Google and France’s competition authority over copyright protections for news snippets, the Autorité de la Concurrence announced a €250 million fine against the tech giant Wednesday (around $270 million at today’s exchange rate).
According to the competition watchdog, Google disregarded some of its previous commitments with news publishers. But the decision is especially notable because it drops something else that’s bang up-to-date -- by latching onto Google’s use of news publishers’ content to train its generative AI model Bard/Gemini.
The competition authority has found fault with Google for failing to notify news publishers of this GenAI use of their copyrighted content. This is in light of earlier commitments Google made which are aimed at ensuring it undertakes fair payment talks with publishers over reuse of their content.
Copyright and competition wrongs
In 2019, the European Union passed a pan-EU digital copyright reform that extended copyright protections to news headlines and snippets. News aggregators, such as Google News, Discover and the “Top Stories” feature box on search results pages, had previously scraped and displayed these news stories on their products without any financial compensation.
Google originally sought to evade the law by switching off Google News in France. But the competition authority quickly stepped in -- finding its unilateral action an abuse of a dominant market position that risked harm to publishers. The intervention essentially forced Google to cut deals with local publishers over content reuse. But in 2021, Google was hit with a $592 million fine after the competition authority found major breaches in its negotiations with local publishers and agencies.
The tech giant called the sanction “disproportionate” and said it would appeal. But it subsequently sought to settle the dispute -- offering a series of pledges and withdrawing its appeal. The commitments were accepted by the French Autorité, include passing key information to publishers and negotiating in a fair way.
Google has signed copyright agreements with hundreds of publishers in France -- which fall under the remit of its agreement with the Autorité. So its business in this area is very tightly regulated.
No appeal
Google has agreed not to contest the Autorité’s latest findings -- in exchange for a fast-tracked process and making a monetary payment.
However, its managing director for news and publishing partnerships, Sulina Connal, struck a peeved tone -- writing in a lengthy blog post that “the fine is not proportionate to issues raised” by the authority.
The blog post suggests Google really wants to draw a line under the saga this time, with Connal also writing: “We’ve settled because it’s time to move on and, as our many agreements with publishers show, we want to focus on the larger goal of sustainable approaches to connecting people with quality content and on working constructively with French publishers.”
With generative AI in the frame, and the competitive scramble to launch tools, Google’s calculus on approaching the content reuse issue looks different.
GenAI training in the frame
Today’s enforcement by France’s competition authority shows it honed in on Google’s use of content from news publishers and agencies for training purposes for its AI foundation model and its related AI chatbot service Bard (now called Gemini).
It found Google used content from publishers and press agencies for training Bard, its generative AI tool which launched in July 2023, “without notifying the copyright holders or the Authority,” per its press release.
On this point, Google’s defense is twofold. In its blog post it writes that the competition authority “does not challenge the way web content is used to improve newer products like generative AI, which is already addressed in Article 4 of the EUCD” [EU Copyright Directive].
Article 4 of the Copyright Directive sets out an “exception or limitation for text and data mining” -- specifically for “reproductions and extractions of lawfully accessible works and other subject matter for the purposes of text and data mining”.
However in its press release the Autorité argues it has not yet been determined whether the exemption applies here. (It’s worth noting the relevant clause refers to “lawfully accessible works” -- while Google is under a legally binding commitment to the competition authority to notify copyright holders about uses of their protected works and apparently failed to do so in this case.)
“When it comes to declaring whether using news content to train an artificial intelligence service falls under neighboring rights and protection, this question has not been answered just yet,” the competition authority wrote. “However, the Autorité considers that Google has breached its commitment #1 by failing to inform publishers that their content had been used to train Bard.”
Google’s blog post also makes passing mention of the EU AI Act -- suggesting it’s of relevance. However the legislation is not yet in force as it’s pending final adoption by the European Council.
The incoming AI legislation will also say developers must abide by the bloc’s copyright rules. And it introduces transparency requirements with that goal in mind -- requiring them to put in place a policy to respect EU copyright law; and make publicly available a “sufficiently detailed summary” of the content used for training general purpose AI models (such as Gemini/Bard).
This incoming requirement on model makers to publish a training data summary may, in the future, make it easier for news publishers whose protected content has been ingested for GenAI training to obtain fair remuneration under EU copyright law.
No technical opt out
The Autorité also points out that Google failed to provide, until at least September 28, 2023, a technical solution to allow publishers and press agencies to opt out of their content being used to train Bard without such a decision affecting the display of their content on other Google services.
“Until this date, publishers and news agencies that wanted to opt out of this use case had to insert an instruction that blocks all content indexation from Google, including for Search, Discover and Google News services. Those services are specifically part of the negotiation for revenue related to neighboring rights,” it wrote, adding: “In the future, the Autorité will carefully look at the effectiveness of Google’s opt-out processes.”
In more technical terms, between July and September 2023, news publishers could insert a “noindex” tag to the robots.txt file to make sure that their content wasn’t used to train Google’s AI model. This robots.txt file is placed at the root folder of web servers and contains various instructions for search engines. Google’s web crawler looks at the instructions in those files to index websites.
But a “noindex” tag means that your website disappears from Google altogether. In September 2023, Google added more granularity and created a “Google-Extended” rule that is different from the “noindex” rule. By opting out of the Google-Extended instruction, web publishers indicate that they don’t want to help improve Gemini’s current and future models.
Other shortcomings
The Autorité is also sanctioning Google for a raft of other issues related to how it negotiates with French news publishers, finding it failed to provide them with all the information needed to ensure fair bargaining of remuneration for their content.
In its press release, it wrote that Google’s information to publishers about its methodology for calculating how much they should be paid was “particularly opaque.”
It also found Google failed to meet non-discrimination criteria, aimed at ensuring publishers get equal treatment. And it called out a decision by Google to impose a “minimum threshold” for remuneration -- i.e. below which it would not make any pay-outs to publishers -- with the Autorité describing this as introducing discrimination between publishers “in its very principle”. Below a certain threshold all publishers are “arbitrarily allocated zero remuneration, regardless of their respective situation”, its press release also noted.
Additionally, the Autorité found fault with Google’s calculations regarding so-called “indirect income”, saying the "package" it proposed was not in accordance with previous decisions or the appeal judgment of the Court of Justice, from October 2020.
It also said Google failed to act on its commitment to update remuneration contracts in line with its pledges.