GenAI Lawsuits: What You Need to Know (and some stuff you don’t)
David Atkinson
AI Legal Counsel | A.I. Ethics and Law | University Lecturer | Veteran
If you want to understand the legal risks of generative AI, you can’t go wrong by first understanding the ongoing litigation (aka lawsuits) against the likes of OpenAI (ChatGPT), Anthropic (Claude), Meta (Llama), Google (Gemini), Perplexity (uh… Perplexity?...), and others.
I’ve written about this kind of analysis before, in May of last year. The full article is here: https://intersectingai.substack.com/p/a-taxonomy-of-legal-risks-for-generative?
Since then, there have been at least 11 new lawsuits against GenAI companies. These include claims around books, songs, chatbot companions and suicide, and more.?
The Claims
What hasn’t changed, however, is that the seven most common claims in initial complaints (the first filings by the plaintiff against the company they are suing) are still the most common claims:
It’s not particularly close, though the order has shifted a bit. There are at least:
Here’s a handy little chart with a description of claims that are most common, as well as ones I identified as possible claims for future litigation made by someone who enjoyed my original paper on this topic.
The above was based on what I thought people might sue over. Below is a more complete table of the actual claims and how many times they’ve been made. Note that I try to use the plaintiffs' verbiage from their complaints, so there may be some overlap and room to quibble. (What do you want from me?) However, it’s directionally accurate.
The next most common claims have each been filed 4 times, and there is a fair long tail. Consider that while there are 7 fair common claims, there are at least 59 unique claims in total, many only made a single time.?
Importantly, many of the initial claims are later dropped or dismissed. This is often because the Copyright Act–the statute used to sue for copyright infringement–is interpreted by many judges as preempting many other claims, like breach of contract, unfair competition, and unjust enrichment.?
The Defendants
The list of defendants is a who’s who among high-valuation GenAI companies and includes or has included the following:
The Stakes
The stakes may be existential for GenAI companies because the penalties can be intense. For example, if a jury finds that the GenAI company infringed “willfully,” the penalty is $150,000 per infringement. The GenAI models are trained on billions of documents, and probably at least 90% of them are copyrighted (songs, code, movies, writings, videos, etc.), so that could add up quickly.?
But even a much smaller number would wreck a GenAI company. For example, the New York Times has alleged (boy does the word “alleged” do a lot of lifting here…) that OpenAI trained on over 10 million NYT articles, for which the NYT registered the copyrights. Math is not my forte (neither is French, but I took two years in high school and the US sprinkles French into our legal jargon like it’s chocolate shavings on a delicious croissant, so I might as well lean into it), but $150,000 x 10,000,000 = a number greater than the GDP of all but 15 of the world’s nations. That is, it’d bankrupt the companies.?
But courts could also force companies to stop making the models available, destroy the models, or do many things in between, few of which would be delights for the companies to comply with.?
The Defenses
Most defenses hinge on a two-step argument:?
I think the fair use argument is bogus in numerous ways (the biggest way is the one most overlooked), ranging from the narrow legal reasoning to the broader policy considerations. But I’ll tackle that in a future post.
Hope you found the post helpful. If you would like me to include anything else, please let me know!