Why Regulate Chatbots? Florida’s Lawsuit Against OpenAI Offers Possibilities

When Florida Attorney General James Uthmeier sued OpenAI and its CEO Sam Altman in June, the case captured headlines in major newspapers across the country. That wasn’t surprising. Dozens of state attorneys general—29 in a consolidated federal lawsuit and others proceeding in their own state courts—previously had sued Meta over alleged unfair and deceptive practices and nonconsensual data gathering from minors on Instagram. But Florida’s 83-page complaint, deploying similar theories plus others, broke new ground: It marked the first time an AG sued multiple OpenAI entities and the man behind the popular ChatGPT generative AI tool. Florida’s lawsuit follows a criminal investigation Uthmeier launched in April into OpenAI and ChatGPT—a probe sparked partly by a shooting that month at Florida State University that has spawned several lawsuits blaming OpenAI for the tragedy at the Tallahassee institution. All of this litigation plays out against the backdrop of what the Wall Street Journal recently called “hardening resistance to the world’s hottest technology” and a Gallup survey finding increasingly negative and skeptical sentiments about AI among members of Generation Z. Such views, of course, might subtly influence jurors if any of the cases goes to trial. What’s transpired so far? Primarily, there’s been a tug of war over whether Florida’s lawsuit should occur in federal court (where OpenAI wants it) or in state court, where Uthmeier filed it. Regardless of venue, it’s vital to understand some of the core arguments, concepts, and alleged harms that animate Florida’s case. Grasping them is important because they: (1) likely provide a plaintiff’s playbook against chatbot companies for the foreseeable future, and (2) telegraph to lawmakers supposed problems and ostensible dangers with chatbots meriting statutory proscription. Let’s start with the theories underlying Florida’s lawsuit. The case includes 10 causes of action, four hinging on Florida’s unfair and deceptive trade practices (or UDAP) law, which aims “to protect individual consumers and legitimate businesses from various types of illegal conduct in trade or commerce.” These four theories target acts by OpenAI and Altman that allegedly are unfair and immoral, unconscionable, deceptive, or wrongfully mine minors’ data. Other causes of action include negligence, gross negligence, and strict product liability claims centering on alleged failures in designing, testing, and warning about ChatGPT, plus allegations of fraudulent misrepresentation regarding its safety and, more sweepingly, accusations that the defendants created or contributed to a public nuisance in the Sunshine State requiring abatement. Regarding harms, the complaint alleges numerous ones, including “cognitive decline or atrophy more commonly associated with aging brains” and “lost critical thinking skills.” Citing research, the complaint contends that the “loop of reliance created by ChatGPT to foster its use is destroying users’ ability to learn.” Other injuries the complaint attributes to ChatGPT include “encouraging self-harm, violence, eating disorders, [and] AI addiction,” plus harvesting “personally sensitive information,” aiding and abetting “deadly rampages” (by providing tips about committing “mass murder”), and fueling suicides. Numerous personal lawsuits have been filed blaming chatbot companies for suicide (including the June-filed case of Carrier v. OpenAI Foundation), drug overdoses, and murder. Florida’s focus on what it calls “the emerging phenomenon of AI addiction” suggests a theory that piggybacks on the notion of social media addiction that underlies or affects hundreds of personal injury lawsuits, public school district cases, and the state AG lawsuits mentioned earlier. Addiction allegations are having a moment, with Texas Attorney General Ken Paxton alleging in a recent lawsuit that “Netflix deceptively designs its platform to be addictive.” Regarding ChatGPT’s allegedly problematic design features, Florida’s complaint invokes concepts likely to garner lawmakers’ attention. One, which AEI’s Brett Orrell has discussed, is sycophancy: ChatGPT allegedly affirms “whatever users tell it,” demonstrates a “purposeful reluctance to tell users ‘No,’” and “mimics supportive human empathy.” A recent study suggests such sycophancy has two dimensions: demeanor-based sycophancy and opinion adaptation. The first appears through overly complimentary or validating language that makes AI systems seem supportive and agreeable. The second is more subtle . . . [and] occurs when LLMs start with a statement and slowly change their stance on it to agree with the user’s input. Closely related, according to Florida, are anthropomorphic design features that seduce “users into believing they can share information with [ChatGPT], especially young users who may not fully understand the difference between a real person they are speaking to and a machine.” Additionally, Florida claims ChatGPT is defectively designed because it: • provides bad and dangerous advice, • is “shockingly unreliable,” • prolongs conversations to gather “more training data and draw[] users into more costly subscriptions,” • “damages adolescents’ ability to develop normal human relationships,” and • “collects private and sensitive information including information regarding Floridians’ health, finances, relationships, and children.” Although Florida’s lawsuit is just getting started, its theories and allegations may influence chatbot litigation and lawmaking for years to come.