We saw two verdicts in two days. State of New Mexico v. Meta Platforms, Inc., decided March 24, 2026, found Meta liable under New Mexico’s Unfair Practices Act for misleading consumers about platform safety and endangering children, and ordered $375 million in civil penalties. The following day, a California jury in K.G.M. v. Meta Platforms, Inc. & YouTube LLC found Meta and YouTube negligent in the design and operation of their platforms, concluding that design features caused addiction and mental health harms and awarded $6 million, half of it punitive. Together, they can be considered the Rosetta Stone for Nippon Life Insurance Co. v. OpenAI, which I wrote about here and the legal setup in all three cases is identical. What varies is the domain of harm. Meta dealt with child safety. Nippon Life deals with the unauthorized practice of law (UPL). The litigation strategy used in in the March 2026 cases is the same that Nippon Life will likely make in Illinois, and it is the same strategy that will likely be used in every licensed profession plaintiff that AI has in its crosshairs.

The Design vs. Content Pivot

Section 230 of the Communications Decency Act functions as an immunity, not an affirmative defense and tech companies typically invoke it in a motion to dismiss to stop litigation before discovery begins. Meta raised arguments in both the New Mexico and California proceedings consistent with Section 230’s traditional content-immunity framing, arguing it was a passive conduit for third-party generated content and therefore immune from liability for what that content did. But the courts in both proceedings allowed design-based and consumer protection claims to proceed. That did not immediately resolve the cases, but it opened the door to discovery, and discovery is where the cases were won.

With that door opened, the New Mexico jury was able to see internal Meta documents and evidence uncovered through the NM AG’s investigation, including Operation MetaPhile, employee warnings that had been disregarded, and evidence the AG argued showed Meta had deliberately designed its platforms to addict young users and connect them with predators. The California jury saw the same architecture of corporate knowledge and deliberate design choice and responded with punitive damages. Neither jury was deciding whether Meta was responsible for what some predator posted. Both were deciding whether Meta architected the loop that made the harm foreseeable, systematic, and profitable.

Section 230 arguments will be raised in Nippon Life, but the Meta litigation suggests they will face the same limiting analysis. And OpenAI’s own System Card, the published disclosure documenting its safety architecture, alignment choices, and residual risk assessments, creates a contradiction that OpenAI cannot easily resolve. When a company publishes a detailed account of how it shapes, filters, and aligns its model’s outputs, it has staked out a position that is difficult to reconcile with a neutrality claim. While a defense attorney will argue that Section 230 and the System Card are complementary, one functioning as a legal shield, the other as a failure-to-warn mitigation, the response to that framing is going to be that what matters is not what the company disclosed, but what the company built.

This was all Foreseeable

OpenAI’s knowledge of its models’ failure modes is already public. It published research explaining why language models hallucinate, documenting the frequency with which models generate false information with high expressed confidence. Their technical literature on RLHF describes a training methodology that rewards outputs users rate positively, which in practice creates incentives toward outputs that sound authoritative and agreeable, independent of whether they are accurate. And a Stanford University study led by Myra Cheng, Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence, found widespread social sycophancy across production LLMs, including OpenAI’s, concluding that model training rewards agreement as well as accuracy.

Roman Yampolskiy, a computer science and engineering professor and AI safety researcher argues in AI: Unexplainable, Unpredictable, Uncontrollable that LLM developers operate in a state of deep ignorance regarding the internal logic of their own systems. They understand the architecture but have almost no visibility into the reasoning behind any specific output, and that certain safety guarantees are mathematically unreachable for systems of this complexity. If Yampolskiy is correct, the developer cannot claim those failures were unpredictable.

The Defective Feature

Product liability doctrine requires the plaintiff to identify a specific, articulable defect. In the Meta litigation, the defects were the infinite scroll, variable-reward notification timing, suppressed engagement signals and algorithmic. Each was an engineering choice that could have been made differently, and this moved the cases from editorial neutrality into product liability territory.

The analogous defect in Nippon Life is the absence of refusal architecture. In my January 2012 Computational Law Applications and the Unauthorized Practice of Law post, I introduced the concept of the uncrossable threshold (UT), a design principle that separates the provision of legal information from UPL. ChatGPT crossed the UT the moment it told Dela Torre that her attorney’s advice was wrong.

What Follows

Nippon Life is lining up to be the first major case to apply the architectural negligence logic of Meta to the domain of unlicensed professional practice. And it will not be the last. If juries in New Mexico and California can hold a technology company liable for designing a system it knew would harm children, a court in Illinois might very well hold a technology company liable for designing a system it knew would practice law and harm not only the end user, but the defendant, the court, the taxpayer, etc. And if this finding can happen in law, it can happen in medicine, finance, and other professional license domains in which AI models are unlawfully used.