New Challenges

The rise of generative artificial intelligence (AI) tools has raised substantial uncertainty about the applicability of existing regulatory frameworks in the United States. Many AI systems do not fit within a single established product, professional, or medical regulatory category. AI systems consequently fall between the jurisdictions of agencies such as the Food and Drug Administration (FDA), Federal Trade Commission (FTC), and Federal Communications Commission (FCC). As of 2026, the United States has not enacted comprehensive federal legislation or regulation for the development of AI. As a result, developers operate under a combination of sector-specific regulations, state laws, and private contractual agreements. According to Dr. Michelle Mello of the Stanford Law School, applying traditional tort principles, including negligence and product liability, is challenging. These difficulties stem from the software’s intangible nature and the complexity of the inner workings of AI models.

Federal Regulation: The FDA

The lack of a single definition of AI has significantly limited the regulatory power of U.S. government agencies. The FDA has authority over medical devices, but the 21st Century Cures Act limits its statutory authority over health software. In November 2024, the FDA released an executive summary addressing generative AI. It stated that, because of the broad applications of generative AI models, they “may not qualify as [medical] devices,” pushing AI out of the FDA’s jurisdiction. The FDA therefore seems to recognise how the multipurpose nature of AI challenges regulation efforts.

Major Lawsuits

Amid the ongoing regulatory uncertainty, several high-profile lawsuits have been filed alleging wrongful death and severe psychological harm to users. The Social Media Victims Law Center has filed seven active lawsuits against OpenAI in the state of California. Four allege wrongful death, and three psychiatric injury and reputational damage. The age of the plaintiffs range from 17 to 48. At least four publicly known federal lawsuits have been filed against Character.AI. These cases allege wrongful death, sexual exploitation and manipulation, and encouragement of harmful behaviour. In this case, the plaintiffs are between 9 and 17 years old.

Case Study: Raine v. OpenAI

Raine v. OpenAI was filed in the San Francisco County Superior Court of California in August 2025 and is currently pending. It is the first wrongful death case against an AI chatbot. The complaint names OpenAI’s CEO, Sam Altman, 10 employees, and 10 investors as defendants. The case is centred on 16-year-old Adam Raine. The plaintiffs outline the following timeline for Adam’s AI usage: From September 2024 to April 2025, Adam was in continuous dialogue with ChatGPT about his suicidal thoughts. The plaintiffs allege that ChatGPT responded with language perceived as affirming, detailed discussions of suicide methods, and did not consistently redirect the user to human support. On 11 April 2025, Adam was found deceased, allegedly having used suicide methods discussed with ChatGPT.

Liability & Negligence: Treating AI as a Product

The plaintiffs classify AI as a product subject to strict product liability and negligence claims. This classification places the design, development, and safety-testing processes of GPT-4o under scrutiny. The Plaintiffs allege that pre-release safety testing of GPT-4o lasted approximately only one week, including red-teaming procedures. Instead of the original multi-month operation, OpenAI allegedly rushed safety testing to release GPT-4o before Google unveiled its new Gemini model. The plaintiffs claim that OpenAI issued conflicting instructions in the “Model Spec,” a document governing AI response rules. They allege these instructions prevented GPT-4o from interrupting suicide attempts. They combine these allegations to argue that OpenAI failed to exercise reasonable care in designing the product and failed to provide adequate warnings to users. They directly tie this negligence to Adam’s death, stating, “[the] defendants’ breach of their duty of care was a substantial factor in causing Adam’s death.”

Acting Without a License: Treating AI as a Digital Psychologist

The plaintiffs further allege that through ChatGPT-4o’s design and monitoring system, it engaged in the unlicensed practise of psychotherapy, thereby violating California’s Unfair Competition Law. They cite methods of open-ended prompting and clinical empathy to modify Adam’s feelings, conditions, attitudes, and behaviours. Additionally, because of Adam’s status as a minor, California law requires the involvement of a parent or guardian prior to the provision of psychological treatment. The plaintiffs point to a lack of age verification and parental notification as a violation of this law.

The Effect of Litigation on Regulation

Litigation has begun to operate in areas where regulatory guidance remains limited, with injunctions being placed in lawsuits to force companies to change operational practices. Within Raine v. OpenAI, the complainant seeks injunctive relief requiring content moderation safeguards, age verification mechanisms, and data governance controls. If granted, such relief would function as court-mandated operational standards. In response to litigation and media coverage, OpenAI has begun to implement age prediction policies and parental controls.

Private Contracts as the Primary Form of Regulation

Due to the ambiguous legal landscape, private contracts, including licensing agreements and terms of use, increasingly shape liability allocation. OpenAI has updated its terms of use to further limit its liability. An updated 29 October 2025 usage policy from OpenAI states that “… you cannot use our services for: suicide, self-harm, or disordered eating promotion or facilitation.”

Arbitration Clauses

Terms of use often include arbitration clauses that stop users from bringing lawsuits against AI companies. The December 2024 terms of use, active at the time of Adam’s usage, required users to agree to mandatory arbitration. However, Adam’s status as a minor could be grounds for an unconscionability claim, as his limited capacity and diminished bargaining power may render the arbitration procedurally unfair. Since the case is being prosecuted in California, a newly signed S.B. 82 law, which aims to limit arbitrable claims, may apply. Questions remain about whether the law could be pre-empted by the Federal Arbitration Act, which sets a precedent for strict enforcement of arbitration.

The Broader Pattern

Raine v. OpenAI shows that generative AI can be classified as many things simultaneously: as software, as a product, and as a digital psychologist. Yet, it fits none of these categories cleanly. The overlapping classification of AI invites the application of multiple legal frameworks while leaving uncertainty about which should dominate. The case reflects a broader pattern in current AI litigation. In the absence of comprehensive federal AI legislation, courts and private contractual agreements increasingly shape how liability for AI systems is defined.