During a long conflict between Google and the French competition authority regarding the copyright protection of news fragments, the Competition Authority ad a fine of 250 million euros against the technology giant.
Google breached some of its previous agreements with news publishers, according to the competition watchdog. However, the decision is exceptionally notable because it includes something else that is very up-to-date: Google's use of news publishers' content to train its Bard/Gemini generative AI model.
Google has been criticized by the competition authority for failing to inform news publishers about GenAI's use of their copyrighted content. This is due to Google's previous commitments to ensure it has fair conversations with publishers about payments for reusing their content.
Errors related to copyright and competition
The pan-European digital copyright reform approved by the European Union in 2019 expanded copyright protection to headlines and news clips. Previously, news aggregators, such as Google News, Discover, and the “Top Stories” feature box on search results pages, had collected and displayed these news stories in their products without receiving any payment.
At first, Google tried to get around the law by removing Google News in France. However, the competition authority quickly intervened and determined that its unilateral action was an abuse of a dominant market position that risked harming publishers. The intervention forced Google to sign agreements with local publishers on reusing content for the most part. However, Google was fined $2021 million in 592 after the competition authority found significant violations in its agreements with local publishers and agencies.
The technology giant declared that it would appeal the sanction, which it described as "disproportionate." However, he attempted to resolve the dispute later, promising several things and withdrawing his appeal. Transmitting crucial information to editors and negotiating fairly were compromises that were accepted by the French authority.
Google and hundreds of French publishers have signed copyright agreements that fall within the scope of their authority agreements. Therefore, your business in this area is highly regulated.
No appeal
Google has reached an agreement not to object to the most recent results of the Authority, in exchange for a faster process and financial compensation.
However, Sulina Connal, the general director of news and editorial partnerships, adopted an irritated tone as she expressed in a lengthy blog post that "the fine is not proportional to the issues raised" by the Authority.
According to the publication, Google intends to end this this time, with Connal stating: "We are happy that it is time to move on and, as demonstrated by our numerous agreements with publishers, we want to focus on broader areas with a sustainable approach to “connecting people with high-quality content and collaborating constructively with French publishers.”
Google's calculus for addressing the issue of content reuse looks different due to the use of generative AI and the existing competition to launch new tools.
GenAI learning in the framework
The French competition authority's current application of the law indicates that it focused on Google's use of content from news agencies and publishers to train its basic AI model and its associated AI chatbot, Bard (now called Gemini ).
According to its report, Google trained Bard, its generative AI tool released in July 2023, “without notifying copyright holders or the Authority.” Press to release.
Right now, Google has two lines of defense. On his blog, he states that the competition authority “does not question how web content is used to improve newer products such as generative AI, which is already addressed in Article 4 of the EUCD” (EU Copyright Directive).
"Reproductions and extractions of works and other legally accessible subject matter for the purposes of text and data extraction" are excluded or limited in Article 4 of the Copyright Directive.
The Authority states in a press release that it has not yet been established whether the exemption applies in this case. Importantly, although Google made a legal commitment to the competition authority to notify copyright holders about the use of their protected works in this case, it did not do so.
The Competition Authority wrote: "This question has not yet been answered when it comes to determining whether the use of information content to train an artificial intelligence service falls within the scope of related rights and protection." However, the Authority believes that Google has breached its number one commitment by not informing publishers that their content had been used to train Bard.
The EU AI Law is mentioned in Google's blog post, indicating that it is relevant. However, the regulations are not yet being implemented because they are pending full approval by the European Council.
Upcoming AI legislation will force developers to comply with the block's copyright regulations. With that goal in mind, transparency requirements are set: they are asked to establish a policy to respect EU copyright law and the public is offered a "sufficiently detailed summary" of the content used to train AI models of general purpose (like Gemini/Bard).
Under EU copyright law, the requirement that model makers publish a summary of training data may make it easier for news publishers whose protected content has been used for GenAI training to receive fair compensation in the future.
There are no technical exclusions
The Authority further notes that Google did not provide a technical solution that would allow publishers and press agencies to opt out of using their content to train Bard without this decision affecting the display of their content on other Google services, at least until September 28, 2023.
Until now, publishers and news agencies that wanted to prevent this use have had to insert an instruction that blocks all Google content indexing, including search, discovery and Google News services. "Those services are specifically part of the negotiation of revenue related to related rights," he wrote, adding: "Going forward, the author will carefully examine Google's effectiveness in opting not to process."
In more technical terms, between July and September 2023, news publishers could add a "noindex" tag to the robots.txt file to ensure that their content is not used to train Google's artificial intelligence model. This robots.txt file is located in the root folder of the web servers and contains a series of instructions for search engines. The instructions in those files for indexing websites are analyzed by Google's web crawler.
However, a “noindex” tag indicates that your website is not indexed by Google. Google added more detail in September 2023 and created a “Google-extended” rule, which differs from the “noindex” rule. Web publishers would thus indicate that they do not want to help improve current and future Gemini models by rejecting Google's extended instructions.
Other defects
The Authority, finding that Google did not provide French news publishers with all the information necessary to ensure fair negotiation of remuneration for their content, is sanctioning Google for a series of other problems related to its way of negotiating with them.
In his lobbying brief, he claimed that the information provided by Google to publishers regarding its method of determining their salary was particularly opaque.
Additionally, Google was found to have failed to meet non-discrimination standards set to ensure all publishers were treated equally. And he criticized Google's decision to set a "minimum threshold" for remuneration, meaning it would not pay publishers, with the authority describing it as introducing discrimination between publishers "in its very principle". The press release states that all editors receive “arbitrarily zero compensation, regardless of their respective status,” if the salary falls below a certain threshold.
Furthermore, the Authority questioned Google's calculations on so-called "indirect revenue", arguing that its "package" did not comply with previous decisions or the Court of Justice's October 2020 appeal ruling.
In addition, he claimed that Google did not fulfill its promise to update compensation contracts.
