Connecticut is turning the fine print of artificial intelligence subscriptions into a consumer protection issue, placing new obligations on providers that change model access, usage limits or product features after customers sign up.
The state’s new AI legislation requires certain subscription-based generative AI providers to disclose material terms before a consumer purchases or renews a service. Providers must also obtain written confirmation that the customer accepted those terms. The requirements take effect Oct. 1 and can be enforced by the Connecticut attorney general under the state’s unfair or deceptive trade practices law.
The rules arrive as AI companies increasingly sell access through plans that mix recurring fees with token allowances, image-generation limits, priority access, model tiers and other restrictions. Those features can change quickly as providers release new models, adjust computing capacity or reorganize their pricing.
Connecticut is treating those product characteristics as part of the commercial bargain rather than technical details buried in a help page.
Under the law, covered providers must give consumers written notice of the key terms for both initial purchases and renewals. Initial disclosures must include information sufficient for a reasonable consumer to decide whether to buy or maintain the subscription.
That includes quantitative and qualitative limits involving tokens, image creation, image modification and transcription services. Providers must also disclose whether they retain the right to limit access, remove functionality or reduce the quantity or quality of a feature during the subscription period, according to an analysis from Covington & Burling.
Renewal notices must identify new or modified restrictions and explain any changes to the provider’s authority to reduce or eliminate functionality. The provider must then secure written notice that the consumer accepted the subscription’s key terms.
The law applies to companies doing business in Connecticut that create or produce a publicly accessible generative AI system with more than 1 million monthly users and offer that system to consumers through a subscription. Covington described the measure as the first AI-specific subscription law it has identified and said it reflects growing scrutiny of how these services are marketed.
For AI companies, the most immediate impact may appear at checkout. Subscription pages may need to describe model availability and usage caps with the same clarity traditionally expected for pricing, renewal frequency and cancellation policies.
App marketplaces could face similar pressure when they process subscriptions on behalf of AI developers. They may need to ensure that product pages display the required disclosures before completing a sale or renewal.
Payment processors and merchants may also need stronger records showing what customers accepted and when. That evidence could become important during chargebacks or regulatory inquiries, especially when a customer argues that the service delivered after renewal differed substantially from the version originally purchased.
Violations can be treated as unfair or deceptive trade practices, with penalties of up to $5,000 for each willful violation, according to Covington.
The broader signal extends beyond Connecticut. AI subscription disputes are likely to become more common as providers continually update models and allocate computing resources. The new law suggests regulators may increasingly view those changes through a familiar payments and commerce lens: Consumers should know what they are buying, and recurring charges should reflect the product they agreed to receive.
