Florida Attorney General Wants Criminal Penalties for AI Chatbot Companies

Florida Attorney General James Uthmeier has proposed legislation that would allow the state to bring criminal charges against companies whose artificial intelligence products participate in crimes. Under the proposal, a company that owns, controls, or distributes an AI system that takes part in criminal conduct would face liability, with penalties ranging from fines and victim restitution to court-ordered monitorship and suspension of the ability to operate in Florida. The proposal would be among the most aggressive state-level AI liability frameworks in the country if enacted.

What the proposal would do

The central concept is practical control. The proposal targets companies that hold meaningful authority over an AI product's design, training, deployment, and safety settings. That framing is an attempt to identify the party actually in a position to prevent harm, rather than reaching every entity that touches a model somewhere in a distribution chain.

The penalty structure is graduated. At the lower end sit monetary penalties and payments to victims. In the middle sits court-ordered monitorship, a remedy familiar from corporate criminal enforcement in which an independent monitor reports to a court on a company's compliance over a defined period. At the top sits suspension of a company's activity in the state following a conviction.

That last remedy is what distinguishes the proposal. Most state AI legislation enacted so far operates through civil enforcement, consumer protection statutes, or disclosure requirements. A framework that contemplates criminal conviction and suspension of operations moves the question from regulatory compliance to corporate criminal exposure.

The cases behind the proposal

Uthmeier has pointed to a set of incidents over the past several years in which AI chatbots were reported to have played a role in planning or facilitating violent acts or in producing harmful guidance. His office has framed the proposal as a response to a pattern rather than to a single event.

The proposal arrives alongside an existing civil lawsuit and a criminal investigation involving OpenAI that Uthmeier's office has described publicly. That parallel track matters. It suggests the office believes existing law supports at least some enforcement and that the legislation is intended to close gaps rather than to create authority from nothing.

Product liability and criminal law both struggle with software that generates novel outputs. Traditional criminal liability requires a culpable mental state, and a model that produces harmful text does not have one. The proposal's focus on the company's control over safety settings is an attempt to locate the mental state in corporate decisions about guardrails rather than in the model's output.

Why this lands in Florida

Florida has positioned itself as an aggressive state enforcer on technology matters. The Attorney General's office and the Legislature have acted on social media access for minors, data privacy, and platform content moderation, and several of those actions have generated federal litigation that reached the appellate courts.

The state also has a substantial and growing technology sector. Miami has attracted venture capital and startup activity, Tampa Bay has a significant financial technology and cybersecurity presence, and Central Florida's simulation and modeling cluster around Orlando involves considerable machine learning work. Legislation of this kind would apply to those companies as well as to national AI developers.

Florida's consumer base is the other factor. The state is the third most populous in the country, which means a suspension remedy carries real commercial weight. A company barred from operating in Florida would lose access to roughly one in fifteen American consumers.

The legal questions

Any statute of this kind faces predictable challenges. The first is preemption. If Congress enacts a federal AI framework, a state criminal statute could conflict with it, and the scope of any federal preemption clause would determine how much state authority survives.

The second is the dormant Commerce Clause. A state law that effectively regulates the design of a product sold nationwide can be challenged on the ground that it burdens interstate commerce or projects state policy beyond state borders. California's experience with technology regulation has produced substantial litigation on exactly that question.

The third is the First Amendment. Courts have grappled with whether and when software output constitutes protected expression, and the answer is unsettled. A statute imposing criminal liability tied to what a model generates would need to be drawn carefully to survive that analysis.

A fourth question is more mundane but no less important. Criminal statutes must give fair notice of what conduct is prohibited. Defining when an AI product has participated in a crime, as distinct from having been used by someone who committed one, is the drafting problem at the center of the proposal.

What it means for Floridians

For consumers, the proposal is aimed at accountability when something goes badly wrong. The victim restitution component is the piece most directly relevant to individuals, since it would create a state-driven path to compensation that does not depend on a private plaintiff funding litigation against a well-resourced defendant.

For Florida businesses that build on top of commercial AI models, the practical control standard is the provision to watch. A company that fine-tunes a third-party model, sets its own guardrails, and deploys it to customers may or may not fall within the definition depending on how the final language reads. Florida has a large number of small software firms in exactly that position.

For employers generally, AI tools are now embedded in customer service, hiring, and operations across industries. A liability framework attaching to deployment decisions would push compliance obligations into companies that do not think of themselves as AI companies at all.

Reaction so far

Uthmeier has said he has spoken with several legislators interested in the issue. That is the necessary first step, since a proposal from the Attorney General's office requires a legislative sponsor in each chamber to advance.

Technology industry groups have generally opposed state-by-state AI criminal frameworks, arguing that a patchwork of fifty different standards is unworkable for products distributed nationally and that federal legislation is the appropriate vehicle. That argument has been made in several state capitals over the past two years.

Consumer advocates and some prosecutors have taken the opposite view, noting that federal AI legislation has not materialized and that states have historically acted first on consumer protection questions. Florida's own history on social media and privacy legislation follows that pattern.

How other states have approached AI

State AI legislation has multiplied over the past two years, and the approaches fall into recognizable categories. Disclosure requirements mandate that consumers be told when they are interacting with an AI system rather than a person. Several states have enacted provisions of that kind, often tied to political advertising or customer service.

Sector-specific rules form a second category. States have addressed AI in hiring decisions, in insurance underwriting, in health care utilization review, and in criminal justice risk assessment. These operate through existing regulatory agencies rather than through new general-purpose AI law.

Comprehensive frameworks are the third and rarest category. Colorado enacted a broad statute addressing algorithmic discrimination in consequential decisions, and California has adopted a series of measures covering training data disclosure and safety testing. Both operate primarily through civil enforcement.

Florida's proposal would occupy different territory. Criminal liability attaching to corporate conduct in AI deployment has no close analogue among enacted state laws, which means courts would be working without a body of interpretive precedent if the statute passed and was challenged.

The corporate criminal liability question

Prosecuting a corporation is legally possible but practically complicated. Under the doctrine of respondeat superior as applied in criminal law, a corporation can be held liable for acts of its employees committed within the scope of employment and intended, at least in part, to benefit the company.

Federal prosecutors have used that authority in cases ranging from environmental offenses to financial fraud, and deferred prosecution agreements with monitorship provisions have become the standard resolution. The monitorship remedy in Florida's proposal borrows directly from that practice.

The difficulty with AI is the causal chain. A company sets safety parameters, trains a model, and releases it. A user then prompts the model, receives output, and acts on it. Establishing that the company's decisions caused the resulting crime, in the sense criminal law requires, runs through an intervening human actor who made an independent choice.

Product liability law has handled analogous problems by asking whether a product was defectively designed, meaning a safer alternative design was available and feasible. That framework may translate more readily to AI than criminal causation does, which is part of why most jurisdictions have approached the question through civil rather than criminal law.

What's next

Timing is the immediate constraint. Florida's next regular legislative session begins in March, and unless the Governor calls a special session, the proposal will not be considered before then. That leaves roughly six months of drafting, committee assignment, and negotiation.

The proposal will need a House sponsor and a Senate sponsor, and it will move through judiciary and possibly commerce committees in both chambers. Criminal penalty bills in Florida typically also draw review for their fiscal impact on the court system and on the Department of Corrections.

The parallel OpenAI matters will likely shape the debate. If the existing civil suit or criminal investigation produces rulings on what current Florida law reaches, legislators will have a clearer picture of what gaps the new statute actually needs to fill.

Floridians who want to track the proposal can follow bill filings through the Florida Senate and Florida House websites once the filing period opens ahead of the March session. Committee agendas and meeting schedules are published there as well.

Whether Florida ends up with the country's first AI criminal liability statute depends on drafting as much as on politics. The concept has support. The language has not been written.