Every generation of technology arrives faster than the tax code can absorb it. Prewritten software took decades to settle. SaaS still has no uniform treatment across the fifty states. Artificial intelligence as a commercial product is moving faster than either, and the compliance implications are landing now, not in three years when guidance catches up.
The fundamental problem is that AI products do not fit cleanly into any category that state tax law was designed to capture. A traditional SaaS subscription involves a customer accessing prewritten software hosted on a vendor's servers. AI products involve a customer accessing a model, a system that generates outputs dynamically based on training data and prompts, which may or may not constitute software in the legal sense depending on whose definition you apply. The customer is not buying a program. They are buying inference, prediction, generation, or automation. States are actively debating whether that is a sale of software, a data processing service, an information service, or an entirely nontaxable professional service, and they are reaching different answers.
For companies selling AI products, this is not a theoretical compliance question. Major AI providers are already collecting sales tax on API usage in states where they have determined a collection obligation exists. The question is whether your product is being taxed correctly in the states where your customers are, and whether your billing system has been configured for a classification framework that is actively evolving. If you have not asked that question yet, the answer is probably sitting in your billing system right now, waiting to be found in an audit.
States do not have AI-specific tax categories. They are applying existing frameworks to new products, and the framework that applies to your AI product depends on how your product is structured and how the relevant state has chosen to classify the closest analog.
Before you can answer whether your AI product is taxable, you need to know which framework a given state uses to evaluate it. That determines everything that follows. The four classification categories that appear most frequently in 2025 and 2026 rulings are prewritten software, data processing services, information services, and nontaxable professional or consulting services. Each carries different taxability consequences. Prewritten software is taxable in roughly half of U.S. states including New York, Pennsylvania, Texas, and Washington. Data processing services are taxable in Texas at 80% of the charge under Tax Code Section 151.0035. Information services occupy a complex middle category in New York, taxable if delivered electronically but exempt if the information is personal or professional in nature. Nontaxable services are exempt in states that do not tax services generally, which covers most states for most service categories.
The classification question for AI is which of these categories the product falls into, and the answer depends on a fact pattern analysis that requires examining what the customer actually receives: access to a software application, the output of a data manipulation process, a tailored informational product, or a service performed by a system that the customer never controls or owns.
The rulings issued in 2025 and early 2026 illustrate the divergence directly and show that identical products can reach opposite conclusions depending on the state.
Indiana issued Revenue Ruling 2025-02-RST in July 2025, addressing the taxability of a generative AI chatbot subscription accessed via web or API. The Indiana Department of Revenue ruled that the service is not subject to sales tax because SaaS is not taxable in Indiana when the customer accesses software remotely without any permanent ownership interest. The ruling extended this logic explicitly to AI: because the AI is accessed electronically with no permanent ownership aspect, it is not subject to sales tax. The true object of the transaction, in Indiana's analysis, is the output generated by the AI, not the software itself.
Kentucky reached the opposite conclusion. Kentucky Tax Facts issued January 20, 2026, states that SaaS is considered a taxable 'prewritten computer access service' in Kentucky, and that this classification applies even with artificial intelligence components. Kentucky is not analyzing whether the AI constitutes software in a traditional sense. It is applying its existing prewritten software access framework to AI-enabled platforms and treating the AI components as part of the taxable access service rather than as a distinct, potentially exempt category.
Illinois presents a split outcome at the state versus local level that is particularly consequential for any company with customers in Chicago. At the Illinois state level, the Department of Revenue examined an AI-powered chatbot in 2025 and ruled that because no software was transferred or owned by the customer, the service is nontaxable under Illinois's general rule that cloud-hosted software accessed without ownership is exempt at the state level. Chicago, however, has imposed its 9% Personal Property Lease Transaction Tax on charges for the use of AI platforms since October 1, 2023. A transaction that is completely nontaxable under Illinois state law carries a 9% Chicago local tax on the same facts.
New York has not issued a specific AI ruling, but the New York Department of Taxation and Finance's existing position under Tax Bulletin ST-128 treats remotely accessed prewritten software as taxable tangible personal property through constructive possession. As confirmed by Moffa Tax Law's 2025 analysis, New York extends this position to AI platforms. If the AI is used primarily to deliver information, it may also fall under New York's taxable information services category. Either way, AI products delivered to New York customers are taxable under existing guidance applied to the technology.
If you sell AI products to Texas customers and have not reviewed your configuration since April 2025, your billing system is probably wrong. Texas does not classify AI as software at all. Instead, it taxes AI inference and related services as data processing under Tax Code Section 151.0035, which covers computerized entry, retrieval, search, compilation, manipulation, or storage of data. Under the Texas Comptroller's April 2025 amendments to Rule 3.330, the scope of taxable data processing services was significantly expanded, and many AI products that were previously treated as nontaxable now fall squarely within it.
The amended rule introduced an ancillary test. If the seller's primary activity constitutes data processing, the entire bundled service charge is taxable as data processing even if other nontaxable components are included. AI inference services, meaning providing predictions, recommendations, or generated content via API, fall within this framework. The tax applies to 80% of the service charge, with a 20% exemption, and the effective combined rate reaches approximately 8.25% when local rates are added. For AI companies bundling inference with professional services or implementation, the Texas rule is clear: the entire charge is taxable unless the AI component is demonstrably ancillary and has no separate value. The burden of proving ancillarity sits with the taxpayer.
Texas's bundling rule adds an important wrinkle for AI companies that bundle inference with professional services, consulting, or implementation. Under the amended Rule 3.330, a data processing service sold for a single charge with another service is not taxable only if the data processing service does not have a separate value and is ancillary to the other service. The burden is on the taxpayer to demonstrate ancillarity. If the AI component is the core deliverable, as it typically is in an AI SaaS product, the entire charge is taxable.
The practical consequence of this divergence is that a single AI product sold to customers across multiple states requires multiple taxability determinations that cannot be resolved by a single classification applied uniformly. An AI chatbot subscription that is nontaxable in Indiana, taxable as prewritten software in Kentucky and New York, taxable as data processing at 80% of the charge in Texas, and taxable at 9% in Chicago but nontaxable at the Illinois state level requires the billing system to classify each transaction based on the customer's location and the applicable state's framework.
AI agents, systems that autonomously execute multi-step business processes, introduce a classification layer beyond subscription chatbots or inference APIs. An AI agent that handles accounts payable, manages customer service queues, or executes procurement workflows is not simply providing software access or generating a text response. It is performing a function that, if performed by a human, would be a nontaxable professional service in most states.
The EverydayCPE analysis of the Indiana ruling makes this question concrete: an AI agent that handles an entire accounts payable function is arguably performing a service, not delivering software. The true object test, which Indiana applied in its chatbot ruling, could support a nontaxable outcome in states that use it. But states that classify by the seller's activity rather than the buyer's intent, as Texas does under its ancillary test, may reach a taxable result regardless of how agentic the workflow is.
No state has issued a ruling specifically addressing AI agents as a distinct product category as of July 2026. The Snell and Wilmer analysis from November 2025 confirms that state and local tax authorities will continue to revisit these issues as AI becomes more embedded in business operations, and that continued administrative and judicial developments will shape how AI is classified across jurisdictions. For companies selling or using agentic AI products, the absence of guidance is not a green light. It is an open classification question that an auditor will resolve using the nearest applicable framework, which may or may not align with how the product is currently being taxed.
The immediate priority is classification, not rate lookup. Most billing systems for AI products were configured either by treating AI as equivalent to standard SaaS, or by applying no tax at all on the assumption that AI outputs are nontaxable services. Both approaches are wrong in a meaningful subset of states, and the state-by-state picture is becoming less favorable to the nontaxable position as more guidance is issued.
The practical steps that define a defensible current position are a state-by-state taxability analysis of your specific AI product, which requires mapping how your product would be classified under each state's existing framework for prewritten software, data processing, information services, and nontaxable services. Texas's data processing framework, Kentucky's prewritten access service classification, and New York's constructive possession standard each require a different fact pattern analysis. A product that is clearly nontaxable in one framework may be clearly taxable in another.
For companies with significant Texas revenue, the April 2025 Rule 3.330 amendments are the most urgent compliance review item. The expanded scope of taxable data processing services and the new bundling ancillary test affect AI products that were previously treated as nontaxable or partially taxable. The effective date of April 2, 2025 means that liability may already be accruing for companies that have not updated their Texas configuration. For companies with Chicago customers, the 9% Personal Property Lease Transaction Tax on AI platforms has been in effect since October 2023, predating most AI compliance reviews. If your billing system is applying Illinois state rates to Chicago transactions involving AI access, it is applying the wrong rate to a separate, higher-rate local obligation.
Does yourAI product have a sales tax problem you have not found yet? State-by-state AI taxability is actively being litigated, ruled on, and revised. The classification your billing system applies today may be wrong in three states by this time next year. CereTax tracks AI and digital service taxability changes in real time, applies the right classification at the transaction level, and gives you the audit trail to defend every decision.
