Not the law. Gen Digital’s US$9,950,000 TCPA fund was finally approved on 14 July 2026 and Hy Cite Enterprises’ US$4,750,000 fund preliminarily approved on 24 March 2026 — both wrong-number “artificial or prerecorded voice” classes, both carrying express denials of liability. Statutory damages are unchanged: $500 per violation under 47 U.S.C. 227(b)(3)(B), which a court may treble at its discretion for a wilful or knowing violation.
Four United States matters are worth reading properly before you rely on any of them. Each is checked here against a primary record — the complaint as filed, or the court’s own order — not the secondary write-ups. Two are settlements: one finally approved, one only preliminarily approved. Two are contested claims that have not been decided, and only one of those two pleads an artificial voice at all. None of the four changed a rule. Everything below was checked on 2 September 2026 against CourtListener’s RECAP dockets, the Federal Communications Commission’s released documents, the eCFR text of 47 CFR 64.1200 current as of 31 August 2026, and the 2024 edition of the United States Code on govinfo.
What 47 U.S.C. 227 actually awards
The two damages figures people quote come from subsections that behave differently. For artificial-or-prerecorded-voice claims, 47 U.S.C. 227(b)(3)(B) allows “an action to recover for actual monetary loss from such a violation, or to receive $500 in damages for each such violation, whichever is greater”. Where the court finds a wilful or knowing violation it may, at its discretion, increase the award to not more than three times that amount — the $1,500 figure. Trebling is discretionary, not automatic.
Do Not Call Registry claims work differently. Section 227(c)(5)(B) caps recovery at up to $500 per violation rather than fixing it, requires more than one call within a 12-month period, and carries an affirmative defence subsection (b) does not: reasonable practices and procedures, established and implemented with due care, to prevent unlawful solicitations. Checked 2 September 2026: neither provision has been amended.
The two approved settlements, read from the orders
Jackson v. Gen Digital Incorporated, No. CV-25-00535-PHX-MTL (D. Ariz.), before Judge Michael T. Liburdi. The final order of 14 July 2026 certifies a class of people called about a LifeLock or Norton account using an artificial or prerecorded voice, on a mobile not assigned to anyone holding such an account, between 19 February 2021 and 30 October 2025. Fund: US$9,950,000, non-reversionary. Seventeen opted out, nobody objected. The order records a forward-looking term — the defendant will use the FCC’s Reassigned Numbers Database — and is explicit about what it is not: “This order, the Agreement, and any and all negotiations, statements, documents, and proceedings in connection with this settlement are not, and will not be construed as, an admission by Defendant of any liability or wrongdoing in this or in any other proceeding.”
Keith v. Hy Cite Enterprises, LLC, No. 3:24-cv-729-jdp (W.D. Wis.), on the public docket. Judge James D. Peterson granted preliminary approval at Dkt. 30 on 24 March 2026, certifying a class of people called with an artificial or prerecorded voice on a mobile not assigned to a Hy Cite customer or accountholder, between 22 October 2020 and 10 September 2025. Fund: US$4,750,000, non-reversionary. The class is defined by that description, not by a headcount: class counsel’s fee motion puts 17,902 as the maximum number of potentially affected telephone numbers, and the settlement agreement estimates the likely number of class members at fewer than 6,500. The order sets an in-person fairness hearing for 6 October 2026, so as at 2 September 2026 this one is not finally approved. The settlement agreement records that “Defendant vigorously denies any wrongdoing or liability related to the allegations included in the Lawsuit and denies any improper conduct or violation of the TCPA.” The class notice tells members: “The Court has not decided who is right or wrong.”
Both classes are defined by who was called — people who were not customers — not by what technology called them. Neither order describes an AI-generated voice.
Mortgage One Funding: the two Michigan suits are not the same case
Two class actions were filed against Mortgage One Funding, LLC in the Eastern District of Michigan within a fortnight of each other in early 2026, and they plead different things. The Lamb complaint, No. 2:26-cv-10450-BRM-APP, filed 9 February 2026 before Judge Brandy R. McMillion, pleads no artificial voice at all. It is a single-count Do Not Call Registry claim under 47 U.S.C. 227(c)(5) over at least four calls and one text between 3 and 17 September 2025, and the only interaction it details is with a named live representative.
The artificial-voice allegations sit in the second suit, filed a fortnight later: Landy v. Mortgage One Funding LLC, No. 2:26-cv-10643-MAG-CI, filed 24 February 2026 before Judge Mark A. Goldsmith. The Landy complaint alleges a single call on 23 January 2026 that opened with an artificial voice saying “I’m calling from the Mortgage One Funding rate team on a recorded line.” On the plaintiff’s account he then answered the prompts, stayed on the line and was transferred to a live person. The count is pleaded under 47 U.S.C. 227(b)(1)(A) and (b)(1)(B) — the artificial-voice prohibition — not the Do Not Call Registry rules. If you are citing a 2026 artificial-voice case against this defendant, it is the Landy docket, not the Lamb one.
Both are contested and undecided. In Lamb the motion to dismiss was fully briefed on 21 April 2026; in Landy the last brief on the motion to dismiss and strike class allegations went in on 3 August 2026. No ruling on either motion appeared on the public docket when checked on 2 September 2026. Nothing has been proved against Mortgage One Funding in either. What Landy is fighting about is not whether AI voice is lawful: the motion argues the plaintiff lacks Article III standing because, on his own pleading, he stayed on the call voluntarily, invoking TransUnion LLC v. Ramirez and Spokeo, Inc. v. Robins. Concrete injury and class definition, not model architecture.
| Matter | Forum / primary record | Status at 2 Sep 2026 | What it turns on | Operational lesson |
|---|---|---|---|---|
| Jackson v. Gen Digital Inc. | D. Ariz., CV-25-00535-PHX-MTL; final order Dkt. 28 | Finally approved 14 Jul 2026; dismissed with prejudice; denial of liability on the face of the order | Artificial or prerecorded voice calls to mobiles held by people who were not account holders | Wrong-number risk, not consent-form risk. The order records the defendant adopting the Reassigned Numbers Database |
| Keith v. Hy Cite Enterprises, LLC | W.D. Wis., 3:24-cv-729-jdp; preliminary approval Dkt. 30 | Preliminary approval 24 Mar 2026; fairness hearing 6 Oct 2026; not finally approved | 17,902 unique numbers the defendant’s own records had designated a potential wrong number — the maximum possible class, not the certified headcount | Your own “wrong number” disposition field is the class list. Make it a suppression trigger |
| Lamb v. Mortgage One Funding, LLC | E.D. Mich., 2:26-cv-10450-BRM-APP; complaint Dkt. 1 | Motion to dismiss briefed 21 Apr 2026; undecided; allegations unproven | Do Not Call Registry, 47 U.S.C. 227(c)(5) — no artificial voice pleaded | Registry scrubbing is a separate control from consent and fails separately |
| Landy v. Mortgage One Funding LLC | E.D. Mich., 2:26-cv-10643-MAG-CI; complaint Dkt. 1 | Motion to dismiss and strike briefed out 3 Aug 2026; undecided; allegations unproven | Article III standing and class definition, after one call on 23 Jan 2026 | One call can carry a nationwide class pleading. Pilot volume is not a risk cap |
The FCC rules that already bound you
The applicable rulings predate all four filings. The FCC’s Declaratory Ruling FCC 24-17 (CG Docket No. 23-362, adopted 2 February 2024, released 8 February 2024) confirmed that the TCPA’s restrictions on “artificial or prerecorded voice” encompass current AI technologies that generate human voices. In the Commission’s words: “As a result, calls that use such technologies fall under the TCPA and the Commission’s implementing rules, and therefore require the prior express consent of the called party to initiate such calls absent an emergency purpose or exemption.” Its ordering clause made the ruling effective on release. Checked 2 September 2026: still in force, not stayed, not vacated, not superseded.
Revocation is where the current position is easiest to get wrong. Under 47 CFR 64.1200(a)(10), a called party may revoke by any reasonable method; “stop”, “quit”, “end”, “revoke”, “opt out”, “cancel” and “unsubscribe” in reply to a text are reasonable per se; requests must be honoured within a reasonable time not exceeding ten business days; and callers may not designate an exclusive means of revoking. All of that is live. The piece that is not is the cross-topic element — where a called party revokes in response to one type of informational message, treating that as applying to all future robocalls and robotexts from that caller on unrelated matters. The Consumer and Governmental Affairs Bureau extended that waiver to 31 January 2027 in Order DA 26-12, released 6 January 2026, and limited it carefully: “We emphasize that this waiver extends only to section 64.1200(a)(10) to the extent discussed herein and does not alter the status quo relating to any other prior Commission rules or rulings addressing revocation of consent.”
There is still no federal rule requiring you to announce that a caller is AI. The proposal to define an “AI-generated call” and require disclosure at consent and at the start of the call remains a notice of proposed rulemaking, FCC 24-84, adopted 7 August and released 8 August 2024 in the same CG Docket No. 23-362; a Federal Register search of that docket on 2 September 2026 returns a proposed rule published 10 September 2024 and no final rule. Drafting implications are in rewriting your consent language for AI calls; the states that do impose disclosure duties are mapped in the state AI call-disclosure patchwork.
What actually reduces exposure
Reassigned and wrong numbers. Both approved settlements are wrong-number classes; neither is about a bad consent form. Query the Reassigned Numbers Database against the date consent was captured, and treat any “wrong number” tag as immediate, permanent suppression across every channel — not a disposition code sitting in a report.
Consent capture and records. Keep the wording shown, the timestamp, the source URL or recording, and the business the consumer agreed to hear from. If it cannot be produced per number, it will not help you. Purchased lists carry their own problem, covered in the consent chain for purchased lead lists.
Revocation handling. Ten business days is the outer limit, not the target. Accept revocation on any channel a reasonable person would use, including mid-call speech, and have the agent’s transcript pipeline write to the same suppression table the dialler reads. Under 64.1200(d)(3), where do-not-call requests are held by someone other than the seller, the seller is liable for failures to honour them.
Registry scrubbing. 64.1200(c)(2)(i) sets out the error safe harbour: written procedures, trained personnel, a maintained internal do-not-call list, and a registry version obtained no more than 31 days before the call. Each element has to exist as a document, not a belief.
Calling windows. 64.1200(c)(1) bars solicitations before 8 a.m. or after 9 p.m. local time at the called party’s location — so resolve the recipient’s timezone at dial time and fail closed when it is unknown.
Disclosure. Even without an AI-specific rule, 64.1200(b) already requires an artificial or prerecorded voice message to state the responsible business’s registered name at the beginning, give a callback number, and — for telemarketing — offer an automated opt-out mechanism within two seconds of that identification. The two-second rule is the one most agent scripts miss.
This article is general information, not legal advice. Get advice from a qualified lawyer in the relevant jurisdiction before relying on any of it.
Frequently asked questions
Did any 2026 settlement decide that AI voice agents breach the TCPA?
No. The Gen Digital final order of 14 July 2026 states: “By entering this order, the Court does not make any determination as to the merits of this Lawsuit, and Defendant denies any liability or wrongdoing.” The Hy Cite agreement records a denial of any TCPA violation. Both class definitions use the statutory phrase “artificial or prerecorded voice”; neither identifies AI-generated voice.
What are TCPA statutory damages in 2026?
Unchanged. Under 47 U.S.C. 227(b)(3)(B) a plaintiff recovers actual loss or $500 per violation, whichever is greater, and a court may increase that to not more than three times the amount for wilful or knowing violations. Do Not Call claims under 227(c)(5)(B) are capped at up to $500 per violation, and carry an affirmative defence for reasonable practices and procedures that subsection (b) does not. Checked 2 September 2026.
Do I have to tell people they are speaking to an AI?
Not under a federal rule as at 2 September 2026 — the FCC’s disclosure proposal, FCC 24-84, is still a notice of proposed rulemaking. But 47 CFR 64.1200(b) already requires an artificial or prerecorded voice message to identify the responsible business at the beginning and provide a callback number, and several US states impose their own disclosure duties.
Has the consent revocation rule taken full effect?
Most of it has. The cross-topic element of 47 CFR 64.1200(a)(10) was waived until 31 January 2027 by Order DA 26-12, released 6 January 2026. The ten-business-day deadline, the per se opt-out keywords and the ban on designating an exclusive revocation method all apply now.
Is a settlement evidence that a company did something wrong?
No, and the orders say so. Settlements are compromises reached to avoid the cost and risk of trial. Both matters here contain express denials of liability, and the Gen Digital order declines to make any finding on the merits. Treat approved settlements as information about where litigation risk concentrates, not as findings of fact.
Does any of this apply to calls made inside Australia?
No. The TCPA is United States law and applies to calls to US numbers. Australian outbound calling runs on a different regime, principally the Do Not Call Register Act 2006 and the Privacy Act 1988.
Consent record first, suppression and reassigned-number checks second, disclosure script third. If you cannot produce a per-number consent record today, you are not ready to dial. Zian AI runs autonomous phone, SMS, email and WhatsApp sales agents and is in partner-application beta. Apply For Partnership.