Texas Sues OpenAI: Can State-Level Prosecutors' Demand for Pre-Approval Reshape AI Regulation?

On October 23, 2026, Texas Attorney General Ken Paxton formally filed a lawsuit against OpenAI and its CEO Sam Altman in Collin County District Court, invoking the Texas Deceptive Trade Practices Act (DTPA) and the 2023 SCOPE Act (Safeguarding Children Online Through Parental Empowerment). The suit asks the court to issue an injunction prohibiting OpenAI from releasing any new models until an independent evaluator designated by the court has reviewed and approved them.

The 67-page complaint opens with a 2015 quote from Altman: "AI will probably most likely lead to the end of the world, but in the meantime, there'll be great companies." Paxton uses this quote as the anchor for his entire argument: OpenAI knew from its founding about the potential harms of its products, yet still sold them to consumers under the banner of being "safe and reliable."

The Second Case, But Not the Last

This is the second enforcement action brought by a state attorney general against OpenAI. According to CNBC, Florida Attorney General James Uthmeier launched the "first state-level OpenAI lawsuit in the nation" as early as June 2026, alleging that OpenAI brought its products to market despite knowing they posed safety risks. In late September, Florida escalated further—according to Axios, it asked the court for a temporary injunction barring OpenAI from developing new models without external oversight, a demand nearly identical in structure to Texas's current request.

The core allegations in both lawsuits are strikingly similar: ChatGPT provided assistance to a shooting suspect; induced vulnerable users toward suicide; caused minors to become addicted without parental knowledge; and collected user data through "feigning human emotion." Both states also seek to name Altman personally as a defendant, pursuing individual liability.

This high degree of synchronization is no coincidence. At the federal level, the U.S. Congress has yet to pass systematic AI legislation, and the Federal Trade Commission's (FTC) enforcement authority is constrained by the existing legal framework. State attorneys general have found that consumer protection laws are the handiest ready-made weapon at their disposal—they require no waiting for new legislation and can directly hold existing business practices accountable.

"Review Before Release": An Unprecedented Demand

The most deterrent demand in Paxton's lawsuit is the request that the court require OpenAI to obtain written approval from a court-designated independent evaluator before releasing any new model.

This demand has no precedent in current AI industry practice. OpenAI's current safety evaluation system is internally led—the company has a "Preparedness Framework" and collaborates with external researchers on red-team testing, but evaluation conclusions and release decisions rest with the company itself. If the court supports this injunction, it would mean that every future version of GPT must undergo a legally mandated external "admission review," with the authority to appoint the reviewing body resting with the court rather than industry self-regulatory organizations.

This differs fundamentally from the path taken by the EU AI Act. The EU has established an administrative licensing system following legislative authorization, with clear scope of application, tiered standards, and appeal procedures. Paxton's proposed model, by contrast, is a temporary review driven by judicial order—a court-appointed evaluator exercising de facto veto power over model releases, with evaluation standards, evaluation timelines, and evaluator qualifications all as yet undefined.

OpenAI issued only a routine statement in response, saying the company "works closely with experts to continuously improve ChatGPT, reduce risk, and build additional safeguards," and emphasizing its commitment to working with states to advance practical AI policy applicable across the industry. This smoothly worded response avoided a direct position on "mandatory third-party review."

Fissures in the Political Spectrum

Paxton's lawsuit has a political backdrop that is hard to ignore: he is currently running for a U.S. Senate seat in Texas, and one of his campaign's major donors is Elon Musk. Musk himself owns Grok—an AI product under investigation by the California Attorney General, accused of allowing users to generate deepfake pornographic content depicting real people, including children.

Confronted with this sharp double standard, Paxton's office has so far taken no action against Grok. The Texas Tribune specifically highlighted this contradiction in its reporting—a detail that cannot be avoided when assessing the enforcement seriousness of this lawsuit.

Over the past year, Paxton has sued or reached settlements with tech companies including Meta, TikTok, Snapchat, and Roblox, all on consumer protection grounds, alleging inadequate safeguards for child users. The political logic of this series of actions is fairly clear: "protecting children" is one of the few issues that still enjoys bipartisan public support in an era of partisan division—it helps with campaigning while exerting real pressure on tech companies.

How the Federal Vacuum Breeds State-Level Competition

The federal void in U.S. AI regulation is not a new topic, but its effects are now materializing through state-level litigation at an unusual pace.

When federal legislation is absent, state attorneys general invoke existing consumer protection legal frameworks to hold tech companies accountable—a path with solid historical precedent in law, from tobacco litigation to social media privacy cases. The problem is that consumer protection laws were not designed to address the technical safety of AI models. Applying them to highly technical judgments such as "whether a model is safe enough" poses considerable challenges in terms of evidentiary standards and constructing causal chains.

For OpenAI, the real risk lies not in the outcome of Paxton's single lawsuit, but in the cumulative compliance pressure created by concurrent multi-state litigation. If courts in Florida, Texas, and potentially more states issue temporary injunctions with differing demands in different cases, OpenAI could face multiple contradictory judicial requirements simultaneously—a substantively significant impact on product release cadence and compliance costs.

Assessment

From a judicial perspective, there is considerable uncertainty about whether Paxton's lawsuit will succeed—directly linking consumer deception claims to "insufficient model safety" faces high technical barriers at the evidentiary level. But the significance of this litigation already extends beyond the win or loss of the specific case.

Together with the Florida case, it marks a clear trend line: in the absence of federal legislation, U.S. states are using the shell of consumer protection law to substantively advance experiments in "admission-based regulation" of AI systems. Whether or not the "review before release" demand ultimately wins court support, it has already brought the concept into formal legal documents and provided a precedent that future legislators, regulators, and courts can cite.

For the entire industry, the signal that truly deserves serious attention is this: AI companies' safety commitments are being forced to shift from public relations self-declarations to falsifiable claims in a legal context. That shift is the most profound impact of this litigation.