BILL
SUBJECT
DATE
The Business Council opposes S.934-A (Gonzalez) / A.3411-B (Vanel) which would require owners, licensees, and operators of generative AI systems to display a conspicuous notice that system outputs "may be inaccurate." While the policy goal of consumer transparency is legitimate, the bill as drafted raises significant concerns related to scope, definitions, proportionality, and workability, as discussed below.
Overbroad Scope — No Risk-of-Harm Threshold
The bill imposes a blanket disclosure requirement on all generative AI systems regardless of whether there is any meaningful risk of consumer harm. Many common Gen AI use cases, like generating an image or creative content, pose no realistic risk of deception or harm from "inaccurate" output. The notice requirement should be limited to contexts where a reasonable risk of harm from deception about the origin or accuracy of AI-generated content actually exists.
We encourage the sponsors to consider adopting the Colorado model, which exempts disclosures where it would be obvious to a reasonable person that they are interacting with an AI system. Colorado requirement language reads:
ON AND AFTER FEBRUARY 1, 2026, AND EXCEPT AS PROVIDED IN SUBSECTION (2) OF THIS SECTION, A DEPLOYER OR OTHER DEVELOPER THAT DEPLOYS, OFFERS, SELLS, LEASES, LICENSES, GIVES, OR OTHERWISE MAKES AVAILABLE AN ARTIFICIAL INTELLIGENCE SYSTEM THAT IS INTENDED TO INTERACT WITH CONSUMERS SHALL ENSURE THE DISCLOSURE TO EACH CONSUMER WHO INTERACTS WITH THE ARTIFICIAL INTELLIGENCE SYSTEM THAT THE CONSUMER IS INTERACTING WITH AN ARTIFICIAL INTELLIGENCE SYSTEM. (2) DISCLOSURE IS NOT REQUIRED UNDER SUBSECTION (1) OF THIS SECTION UNDER CIRCUMSTANCES IN WHICH IT WOULD BE OBVIOUS TO A REASONABLE PERSON THAT THE PERSON IS INTERACTING WITH AN ARTIFICIAL INTELLIGENCE SYSTEM.
"May Be Inaccurate" — Vague and Overbroad Warning Language
Requiring a notice that “outputs may be inaccurate or inappropriate" is an ill-fitting standard for many Gen AI applications where accuracy in the traditional sense is irrelevant (e.g., image generation, creative writing tools, internal workflow automation). The bill does not define what constitutes "inaccuracy" or “inappropriate” in these contexts, nor does it distinguish between categories of use. This creates compliance ambiguity and may ultimately mislead consumers by attaching uniform warnings to interactions where the concept of accuracy does not meaningfully apply.
Internal Business Use Should Be Excluded
As currently written, the bill applies to Gen AI deployed both externally and internally within a business. Requiring inaccuracy disclosures on every internal business chatbot would create an enormous and impractical compliance burden. Internal tools typically have no consumer-facing risk profile. The bill should be explicitly limited to consumer-facing Gen AI systems, with a clear definition of what constitutes a "consumer" interaction.
Penalty Structure Is Disproportionate and Harm-Agnostic
The $1,000 civil penalty per user per violation is assessed regardless of whether any actual harm resulted from the absence of notice. This structure creates potentially catastrophic exposure for large-scale deployments, even in cases of good-faith compliance failures or where consumers suffered no demonstrable injury. Penalties should be tied to actual harm caused, or at minimum subject to a safe harbor for good-faith compliance efforts and cure periods.
Definitions of AI and Generative AI Require Refinement
The bill's definition of "generative artificial intelligence" is technically imprecise in two respects. First, the requirement that systems be "self-supervised" excludes common hybrid and supervised-learning models and creates uncertainty for operators who may not have visibility into the underlying architecture of third-party models they deploy. Second, the phrase "derived synthetic content" may not capture retrieval-augmented generation (RAG) systems that surface existing content rather than generating novel outputs. Cleaner, industry-aligned definitions would improve clarity and workability.
"Clearly and Conspicuously" — Ambiguous Display Standard
The bill requires disclosures to be displayed "clearly and conspicuously" but provides no safe harbor or specification for what that standard requires in practice across varied user interface contexts (mobile apps, embedded APIs, voice interfaces, etc.). Without objective criteria, operators face legal uncertainty and potential enforcement risk even when making good-faith disclosure efforts. The bill should either define this standard with specificity or provide examples of compliant implementation.
Without necessary changes to scope, definitions, proportionality, and workability, The Business Council must oppose S.934-A (Gonzalez) / A.3411-B (Vanel).