The Five Highest-Risk States for AI Voice Compliance
State AI voice law is not uniform, and treating compliance as one nationwide setting is how companies get caught. And the state AI voice laws are changing fast, even though the Trump Administration is attempting to create federal guidelines. Most AI voice companies think about compliance the way they think about weather — something that happens everywhere, roughly the same, maybe worse in some seasons. That's the wrong model.
Five states account for the overwhelming majority of class action exposure, regulatory enforcement risk, and per-violation penalty potential you will face when you deploy an AI voice agent in the United States. The other 45 exist on a spectrum from “manageable federal TCPA compliance” to “moderate state-specific layers.” These five are in a different category.
AI Voice Compliance Risk by State
| State | Primary Risk | Key Statutes | Private Right of Action |
|---|---|---|---|
| 1California | CIPA wiretapping class actions | CIPA, CCPA/CPRA, AB 2905 | YesCIPA |
| 2Illinois | BIPA voiceprints; all-party eavesdropping (felony) | BIPA, Eavesdropping Act, ATDA, Consumer Fraud, Right of Publicity | YesBIPA |
| 3Texas | AG biometric enforcement at scale | CUBI, TDPSA, TRAIGA, Tel. Solicitation Act (SB 140) | NoAG only |
| 4Washington | Dual biometric frameworks + autodialer ban | RCW 19.375, MHMDA, RCW 80.36.400 | YesMHMDA |
| 5New York | Right of publicity + UDAP | Civil Rights Law 50/51/50-f, S.8420A, FAIR Act (GBL 349) | Yes |
Ranked by composite score from the Henson Legal 50-State AI Voice Law Survey.
1. California AI Voice Law: CIPA, CCPA/CPRA & AB 2905
California is not the hardest AI voice state by a hair. It is the hardest by a significant margin over most of the country, and only roughly tied with the next tier in raw legal complexity.
The exposure is not theoretical. Plaintiffs' attorneys are filing cases right now. The theory is the California Invasion of Privacy Act is a highly litigated statute. And part of CIPA is AI wiretapping statute which would require consent for the call to be recorded. Statutory damages are $5,000 per violation. Per call. No actual harm required.
Then layer on top:
• CCPA/CPRA classifies voiceprints as sensitive personal information. If your AI voice system creates or processes a voiceprint identifier, you are handling sensitive PI — which means opt-in consent requirements and data minimization obligations before your first call goes out.
• AB 2905 (effective January 1, 2025) requires disclosure on any call using an automatic dialing and announcing device which also uses an AI-generated or AI-altered voice. This is separate from the TCPA's prerecorded-message disclosure. It's California-specific and carries its own penalty structure.
• Dual enforcement. The California Privacy Protection Agency and the AG both have enforcement authority, and the CPPA has explicitly identified AI as an enforcement priority.
Bottom line: You cannot run an outbound AI voice campaign in California without a CIPA analysis, a CCPA/CPRA sensitive-PI protocol, and an AB 2905-compliant disclosure architecture. Do all three and you can operate. Miss any one and you're not compliant — you're just not caught yet.
2. Illinois AI Voice Law: BIPA Voiceprints & All-Party Eavesdropping
Illinois has something California doesn't: BIPA.
The Biometric Information Privacy Act explicitly covers voiceprints. That word — voiceprint — appears in the statute. If your AI voice platform captures, processes, or stores a voiceprint, even temporarily during a call, you need written notice, a written release, a disclosed retention period, and a published destruction policy before you can lawfully do it.
The private right of action is live. Statutory damages run $1,000 per negligent violation and $5,000 per intentional or reckless violation, with no actual injury required for standing. The Fireflies.AI class action claimed that Fireflies (the recording software) collects the voiceprints of meeting participants via the Fireflies software without providing proper written notice. This class action lawsuit shows what BIPA enforcement looks like at scale.
Then there's the eavesdropping statute. Illinois is also an all-party consent state, and violating it while using an AI voice system is a Class 3 felony — criminal liability, not a fine. If you're recording or monitoring a call with an Illinois resident without affirmative consent from all parties, you're in criminal territory.
One AI voice call in Illinois could simultaneously trigger BIPA, the eavesdropping statute, the Automatic Telephone Dialers Act (regulating the use of “autodialers”), the Consumer Fraud Act, and the Right of Publicity Act (prohibiting the distribution of a “unauthorized digital replica” of another’s voice/image/likeness). Each call could end up with five independent causes of action from a single noncompliant call.
Bottom line: Illinois is the highest litigation-risk state in the country for AI voice. California has more enforcement volume; Illinois has more plaintiffs'-bar leverage per call.
3. Texas AI Voice Law: CUBI, TDPSA, TRAIGA & SB 140
Texas surprises people on this list. It shouldn't.
CUBI — the Capture or Use of Biometric Identifier Act — explicitly covers voiceprints. The civil penalty is $25,000 per violation. There's no private right of action; the AG enforces it exclusively. That might sound like good news. It isn't.
Texas has produced two of the largest biometric privacy settlements in the country, both driven by a single AG office with no private plaintiffs needed. The Texas Data Privacy and Security Act (TDPSA, effective July 1, 2024) classifies biometric data processed for identification as sensitive data, which requires affirmative consent.
TRAIGA — the Texas Responsible AI Governance Act, effective January 1, 2026 — adds AI governance requirements and prohibits using AI systems to manipulate consumer behavior through materially deceptive means. The Texas Telephone Solicitation Act, expanded by SB 140 (effective September 1, 2025), adds registration requirements and private rights of action. We are seeing the Texas Telephone Solicitation Act being used as an “add-on” to TCPA claims already.
Bottom line: Texas is the highest-stakes enforcement state for biometric violations. No private plaintiffs — but the AG doesn't need them, and the track record proves it. But, the Texas Telephone Solicitation Act does have a private right of action. And it’s being used already against companies.
4. Washington AI Voice Law: RCW 19.375, MHMDA & the Autodialer Ban
Washington has two separate biometric frameworks covering voiceprints — and an autodialer ban most companies don't know exists.
• RCW 19.375 covers voiceprints explicitly as biometric identifiers, requiring notice and consent before enrollment for commercial purposes.
• The My Health My Data Act (MHMDA) covers voice recordings from which biometric identifier templates can be extracted — and does not require that you actually extract the identifier. The risk attaches at the recording level. MHMDA has a private right of action, and violations are per se violations of the Washington Consumer Protection Act, which carries treble damages.
• RCW 80.36.400 bans unsolicited commercial solicitation using an “automatic dialing and announcing device” which would include MOST AI voice platforms. Not limits — bans. It's a flat prohibition on marketing calls without consent.
Washington also enacted a wave of AI legislation in 2025–2026: HB 1205 (forged digital likenesses, effective July 27, 2025), HB 1170 (AI content watermarking, effective January 1, 2028), and HB 2225 (AI companion chatbot disclosure, effective January 1, 2027). Governor Bob Ferguson — a former AG — has made AI regulation a signature priority.
Bottom line: Two biometric frameworks, one autodialer ban, and an ex-AG in the governor's office. Washington is a state that will be tested early.
5. New York AI Voice Law: Right of Publicity, S.8420A & the FAIR Act
New York's risk profile is different from the other four. It's not primarily about biometrics or wiretapping — it's about right of publicity and the sheer scope of enforcement infrastructure.
New York has the nation's strongest statutory right of publicity framework, in Sections 50 and 51 of the Civil Rights Law. Using a synthetic voice that sounds like a real, identifiable person sits squarely within the statute, with damages covering actual damages plus profits plus punitive damages. New York also enacted expanded digital replica protections for deceased performers in December 2025, eliminating the disclaimer carveout and removing the “likely to deceive” requirement that had limited prior exposure.
S.8420A — the Synthetic Performer Disclosure Act — requires conspicuous disclosure when an advertisement features an AI-generated voice performance, effective June 9, 2026. The FAIR Business Practices Act amended GBL 349 to add “unfair” and “abusive” to New York's consumer protection statute — its most significant expansion in 45 years. An AI voice agent that sounds human without disclosure now has a stronger UDAP exposure vector in New York than it did two years ago.
AG Letitia James has made AI enforcement a public priority and leads multistate coalitions. NYC adds a separate layer through Local Law 3 (biometric) and Local Law 144 (AI in hiring).
Bottom line: New York's right of publicity framework is the sharpest edge for AI voice companies cloning real people. For synthetic AI personas, the UDAP and disclosure exposure is the primary risk.
What This Means Operationally
If you're deploying an AI voice agent, these five states are not places where you adjust your script. They're places where the entire compliance architecture — consent language, call flow, data handling, vendor contracts — has to be built around the specific statute.
The right way to think about it: California and Illinois are your two-party-consent plus biometric-exposure states. Texas is your AG-enforcement state. Washington combines biometric law, an autodialer ban, and fast-moving legislation. New York is your right-of-publicity and UDAP state. Each requires a different legal analysis. They can't be covered by the same consent disclosure.
Frequently Asked Questions
Which states have the strictest AI voice laws?
California, Illinois, Texas, Washington, and New York carry the highest combined litigation and enforcement risk for AI voice. California leads on enforcement volume and CIPA wiretapping exposure; Illinois carries the most per-call litigation risk through BIPA and its all-party eavesdropping statute.
Is it legal to use an AI voice in California?
Yes, but only with the right architecture. An outbound AI voice campaign in California needs a CIPA analysis, a CCPA/CPRA sensitive-personal-information protocol for voiceprints, and an AB 2905-compliant disclosure that the voice is AI-generated. Missing any one of the three creates exposure.
Does BIPA cover AI voiceprints?
Yes. Illinois's Biometric Information Privacy Act names voiceprints explicitly. If your platform captures, processes, or stores a voiceprint — even temporarily during a call — you need written notice, a written release, a disclosed retention period, and a published destruction policy first. BIPA has a private right of action with statutory damages of $1,000 to $5,000 per violation.
Do I need consent to record calls with an AI voice agent?
In all-party-consent states like Illinois and California, yes — from every party. In Illinois, recording or monitoring without all-party consent can be a Class 3 felony when an AI voice system is involved. Consent requirements vary by state, which is why a single nationwide disclosure rarely works.
How do these state laws interact with the TCPA?
State AI voice laws layer on top of the federal TCPA — they don't replace it. The 11th Circuit struck down the FCC's one-to-one consent rule, removing one federal layer, but state-level consent and biometric regimes (like California's) generally remain intact and are not preempted. You have to comply with both.
