Hi, Luiza Jarovsky here. Welcome to the 193rd edition of my newsletter, read by 58,700+ subscribers worldwide. It's great to have you here!
Paid subscribers never miss my full analyses of AI's legal and ethical challenges, dissecting the latest developments in AI governance (3 to 4 times a week):
For more: 4-Week Training | Learning Center | Forum | Job Board
🦙 Meta’s AI in Legal Trouble
In the past few days, Meta's AI practices have been legally challenged from a copyright perspective and in relation to the open-source status of Llama, one of its AI models.
These are two major issues. Although they originate in the EU and the U.S., they will directly affect how Meta is regulated and scrutinized globally and whether its AI business model will survive in its current form.
Copyright
Last week, various prominent groups wrote amicus briefs against Meta in the context of one of the most significant AI copyright lawsuits filed against the company (by authors Richard Kadrey, Sarah Silverman, and others in 2023).
The briefs urge the court to reject Meta's claim that its use of copyrighted works to train AI qualifies as fair use. Relevant quotes from the briefs:
Copyright Law Professors
“Meta’s use of plaintiffs’ works to train its LLM is a commercial use that takes the expression in those works, that is not transformative, and that does not qualify as the kind of ‘intermediate copying’ that courts have found to be a fair use. Meta’s purpose is to create a model that will produce outputs that compete with the works on which it was trained, and with their authors. (…). Taken together, these factors weigh conclusively against a finding of fair use.”
The International Association of Scientific, Technical & Medical Publishers
“Meta knowingly copied and distributed a shocking amount of infringing material from the world’s most notorious infringing websites to serve its commercial ends. Meta’s brazen acts of infringement, unprecedented in the annals of copyright law, must be considered in the context of fair use and should weigh heavily against it.”
The Copyright Alliance
“The cases relied on by Meta for its fair use arguments are of limited relevance here given the narrowed interpretation of transformativeness post-Warhol. This Court should deny Meta’s motion for summary judgment.”
The Association of American Publishers
“In weighing Meta’s claim of fair use, this Court can and should consider Meta’s conduct in relation to the objectives of the Copyright Act and Constitution’s directive to protect the exclusive rights of authors in their works. In addition to avoiding the inconvenience and expense of licensing and compensating copyright owners for the commercial use of their content, Meta opted to evade technological protections that are essential to a functioning online marketplace for copyrighted works. (…) A finding of fair use in this case would not only undermine the public interest in a workable copyright regime, but encourage and reward theft twice over.”
Why do these amicus briefs matter?
Meta claims that it's fair use to train AI on copyrighted works. If the court rejects this claim (in this and other lawsuits), there will be direct implications for Meta's AI practices, which might include the obligation to:
Compensate authors whose work was used to train AI
Negotiate licensing agreements with copyright holders for all future AI training
(To learn more about AI and copyright issues, read my deep dive Can AI Ignore Copyright?)
These would be meaningful changes to Meta's AI business model, especially if other jurisdictions follow suit. It's unclear whether Meta is prepared for them or what the practical consequences might be.
LLaMA's Open-Source Status
Meta's second legal trouble has to do with the open-source status of LLaMA, one of its AI models: