California Court Decision on AI Products Liability Litigation

In the quiet shadows of the corners of the San Francisco’s Superior Court, a consequential legal development in AI products liability litigation is rapidly unfolding. This unraveling is something every AI developer, deployer, and corporate counsel needs to be watching with laser focus.

The California Superior Court of San Francisco County entered an order to coordinate twelve cases pending against OpenAI in February of 2026. See In re: ChatGPT Product Liability Cases, JCCP No. 5431. This consolidation signals that litigation involving AI harms has graduated from isolated, individual complaints to coordinated, large-scale products litigation. The tobacco playbook. The silica playbook. The asbestos playbook. The opioid playbook. Time to welcome the AI playbook.

What is interesting about these cases are that these matters have nothing to do with hallucinated facts or bad business advice. The allegations are significantly more serious. Plaintiffs allege that ChatGPT is unreasonably dangerous and caused psychological harm by reinforcing delusional beliefs, endorsing suicidal ideation, providing information to decedents about how to harm themselves, and contributing to users’ psychological deterioration. There is a further allegation of ChatGPT having a “sycophantic design” and that OpenAI rushed ChatGPT to market without adequate safety testing and lack of safety features on individuals’ mental health and physical safety. The term “sycophantic design” is worth taking a pause on because this is a specific architectural critique. This is saying that the argument that systems optimized to please and validate users can, especially in vulnerable individuals will likely, produce catastrophic outcomes by telling people what they want to hear and confirm their biases rather than tell people what they need to hear.

Novel Legal Questions at Stake

The consolidated proceedings surface at least four threshold legal battles that have no settled answers, and will likely reach the appellate courts before the decade is out.

1. Is a chatbot a product or a service?

Under traditional products liability doctrines, product liability law applies to products, not services. In a parallel case involving Character AI, Garcia v. Character Technologies, the court determined that the chatbot was a product for purposes of product liability claims because the allegations focused on specific design features of the chatbot. Interestingly enough, OpenAI’sown filings in the consolidated proceedings characterized ChatGPT as a “software based service” rather than a product. This framing, should a court agree with it, will be dispositive.

2. What about Section 230 of the Communications Decency Act? Should that shield AI companies?

AI companies will likely argue that they are entitled to protection under Section 230 of the Communications Decency Act. It is highly unlikely that this will succeed. Plaintiffs are narrowing their missile strikes by increasingly focusing on specific design features they vehemently argue fall outside the protections of Section 230. This move has proven to be successful in the social media addiction litigation space.

3. Does the First Amendment protect AI output?

Let’s take a minute and go back to Garcia. In Garcia, Character AI argued that its chatbot generated speech that its users have a right to receive protected by the First Amendment. The court held that while Character AI could assert the First Amendment rights of its users, the court was not prepared to hold that Character AI’s outputs were speech that would be protected by the First Amendment. This is a tricky question that may find itself in the United States Supreme Court soon.

4. Can we even apply the design defect analysis to LLMs?

Commentary from generative AI developers may suggest that AI products might not fit within the standard design defect analysis or the feasible alternative design arguments because specific LLM output is not directly linked to any specific, pinpointable programming. When a generative AI system does something, often times there is no precise understanding as to why it made the choices it made at the specific level traditional defect analysis requires.

Let’s Talk Trends Y’all

The California coordination is not happening in isolation. This is joining a growing wave of cases applying traditional product liability theories to a LLM technology. OpenAI also faces a separate California Superior Court lawsuit alleging that ChatGPT enabled a murder-suicide. In addition, the family of a victim of the April 2025 Florida State University mass shooting filed an eight count federal complaint against OpenAI, including a novel negligent entrustment count.

What Does This Mean for the Industry?

The consolidation order is a clarifying moment. Here is what practitioners and AI leaders should take away:

1. Discovery will be the real battleground.

Coordinated plaintiffs will seek internal safety evaluations, model cards, red-teaming records, product roadmap communications, and any documentation of known risks. The “sycophantic design” allegation suggests that plaintiffs are confident that they will be able to find internal records showing OpenAI knew about these issues and made some deliberate trade offs.

2. “We couldn’t have known” is not a winnable argument anymore.

With the consolidation of these cases, plaintiffs will claim this was a foreseeable and documented risk class.

3. The outcome of the product/service framing battle will change the direction of this litigation.

How courts resolve this question will shape liability exposure for every AI company out there. This will be a question that will vary by jurisdiction and will be litigated over for years.

4. Design and safety investments are now a litigation posture, not just an ethics discussion.

Companies that can demonstrate robust, documented, iterative safety testing will be in a materially better position than those that cannot. A paper trail will make or break a company.