Is AI Cold Calling Legal? What the FCC Ruling and the 2026 Cases Say

The ruling is FCC 24-17, a declaratory ruling dated 8 February 2024. It covers calls that use AI technologies to generate human voices, including voice cloning. It did not create a new rule. It confirmed that an old one applies.

The old rule is in 47 U.S.C. 227. It bars calls that use an artificial or prerecorded voice to a wireless number without the prior express consent of the called party, and it bars such calls to residential lines without consent unless an exemption applies.

An interactive agent that listens and answers is still producing a synthetic voice. Nothing in the sources we reviewed treats a conversational agent differently from a recording for this purpose.

The FCC's rule at 47 CFR 64.1200 defines prior express written consent as a signed written agreement. It must clearly disclose that the person authorizes telemarketing calls using an autodialer or an artificial or prerecorded voice to a specified number, and that agreeing is not a condition of purchase.

There is one current wrinkle. On 25 February 2026 a Fifth Circuit panel held in Bradford v. Sovereign Pest Control that the statute itself does not require the consent to be written, so oral consent can be enough in that circuit. Duane Morris summarized the ruling on 2 March 2026. It covers Texas, Louisiana and Mississippi, and the FCC rule remains on the books elsewhere.

Even under Bradford, consent is still required. The ruling changes the form it may take in three states. It does not make a call to a stranger lawful. Which standard applies to your calls is a question for counsel.

In February 2026 a putative class action was filed in federal court in Michigan against a mortgage lender over alleged AI-generated cold calls, as reported by Inman. On 10 July 2026 another was filed in Texas federal court against mass-tort law firms over AI-agent solicitation calls, under both the federal statute and Texas law, as described by Sheppard Mullin.

The settlement that drew attention came in November. TCPAWorld, a blog run by a defense law firm, reported on 9 November 2026 that Finley v. Altrua settled for 1.1 million dollars. By its account the calls reached about 900 people across nine days and arose from a vendor's test run. That works out to roughly 1,200 dollars for each person in the class.

The statute allows 500 dollars per violation and up to 1,500 dollars when the violation is willful or knowing, with no cap on the total for a class. A short test can therefore carry a large bill.

In Lowrey v. Twilio and OpenAI, in the Western District of Virginia, the court ruled on 1 October 2026 that the complaint adequately alleged the two companies made and initiated the calls. The law firm Rosenberger + Kawabata summarized the ruling on 8 October 2026. We have read summaries of the ruling and have not read the opinion.

Two limits matter. The ruling is at the pleading stage, so the court assumed the complaint's facts were true and found nothing about liability. The court also dismissed the do-not-call claims that were based on texts.

For a business using these platforms, the ruling does not move responsibility away from the caller. The business that directs the calls is still the first defendant.

Retell's documentation for its compliance checks, as of October 2026, lists checks for do-not-call registries, reassigned numbers and known litigators. It also states the limits. Results are a snapshot and are not refreshed before later attempts. The documentation says that 'if the provider can't determine whether a number matches a check, the number passes'. Retell does not enforce calling hours, and consent is entirely the customer's responsibility.

Bland publishes a post saying its product cannot be used for cold-calling campaigns.

None of this is a criticism of the platforms. A list check can tell you a number is on a registry. It cannot tell you whether this person agreed to hear from this seller. Only your own records can.

• Checks are run once and can be out of date by the time of a later attempt
• An inconclusive check lets the number through
• Calling hours are not enforced by the platform
• Proof of consent is yours to hold and produce

The FCC's rules for artificial-voice calls ask for more than consent. The caller must be identified at the start of the message, and telemarketing calls need an automated way to opt out during the call. Telephone solicitations are barred before 8 a.m. and after 9 p.m. at the called person's location.

The list below is drawn from those rules and from what the consent definition requires. It is a starting point for a conversation with counsel, since state laws can add to it.

• The exact consent wording the person saw, with the date and time
• The phone number the consent names, and the seller it names
• A statement in that wording that agreeing is not a condition of purchase
• Where the consent came from: your own form, or a named lead source
• The opening line in which the agent identifies the caller
• A working way to opt out during the call, and a record of each opt-out
• The called person's local time, and the basis you used to work it out
• A note on each call that an AI voice was used