Four weeks before the midterms, the fight over AI robocalls has reopened, and the likely winner is not the voter. On September 24, 2026, the National Consumer Law Center flagged that the FCC is taking public comment on a petition from the Club for Growth that would let political robocalls, including ones using an AI-generated voice, reach your cellphone without your consent. That consent rule is the main thing keeping legal robocalls off your phone. If it goes, the guardrails states spent two years building get tested in the worst possible month.
Here is the short version up front: AI voices are already illegal in most robocalls, states have passed a patchwork of election deepfake laws, and the newest proposal is an attempt to punch a political hole through all of it. The technology that made one fake call in 2024 cost about a dollar and twenty minutes. Now imagine it at the scale of a national election.

The news: a petition to drop consent, one month before the vote
The Club for Growth has asked the FCC for an emergency waiver that would allow “AI-powered calling tools” to contact voters on their cellphones without first getting consent. The petition proposes limits: three calls per 30-day period per caller, and a stop within 10 days of someone asking to be left alone. The FCC has opened the request for comment (proceeding 02-278, DA-26-940A1).
Consumer advocates call those limits meaningless. Patrick Crotty, an attorney with the National Consumer Law Center, told Gray News in early October 2026 that removing the consent requirement could increase political robocalls by as much as 40 times compared with previous midterms, pushing the total toward “15 billion robocalls.” Three calls per committee per month sounds small until you count the committees. And a 10-day stop window means any caller can keep dialing straight through the final stretch of a campaign.
The Club’s argument is not absurd. It says AI calling tools would let campaigns reach voters in more languages and “more efficiently,” and that similar rules already apply to landlines. But “more efficiently” is doing a lot of work here. The same tool that makes a call cheap to place makes a lie cheap to scale.
The one time this already worked
We do not have to guess how this plays out. In January 2024, on the eve of the New Hampshire primary, voters received a robocall that sounded like President Joe Biden telling them to “save” their votes and stay home. It was the first known use of voice-cloning AI at significant scale to sway a US election.
The specifics are the whole story. The Brennan Center reported that creating the fake audio cost roughly one dollar and took under 20 minutes. The call hit thousands of voters two days before the vote. Investigators traced it to a political consultant, Steve Kramer, who had worked for a rival primary campaign.
The consequences were real, and so were the limits:
- The FCC issued a 6 million dollar fine against Kramer in September 2024.
- New Hampshire charged him with 13 felony and 13 misdemeanor counts.
- Lingo Telecom, the provider that transmitted the calls, paid a 1 million dollar settlement and accepted a first-of-its-kind compliance plan.
- A civil suit ordered Kramer to pay 22,500 dollars to three voters. In November 2025, he said he would not.
That last line is the honest summary of enforcement: even when the system works, collecting is a different problem entirely.

What the law actually says right now
The base layer is a 1991 statute. The Telephone Consumer Protection Act requires callers to get prior express consent before placing automated calls to cellphones. For years, the open question was whether an AI-cloned voice even counted as “artificial.”
On February 8, 2024, the FCC settled that with a Declaratory Ruling: calls made with AI-generated voices are “artificial” under the TCPA, making voice cloning in robocalls illegal without consent in many cases. The ceiling matters too: penalties can exceed 23,000 dollars per call, and the ruling opened the door to private lawsuits and state enforcement.
On top of that federal floor, states built their own regimes. In its June 2026 count, the National Conference of State Legislatures tallied 31 states with election deepfake laws. Of those, 28 require disclosure. The other three, Maryland, Minnesota, and Texas, prohibit certain deceptive deepfakes even when the content carries a warning.
Two 2026 laws show how differently states read the problem:
- Louisiana passed a May 2026 law requiring robocalls that use AI to reproduce a public figure’s voice to disclose that use at the start of the call. Civil fines reach 2,500 dollars for a first violation and 5,000 dollars after.
- Maryland passed a May 2026 law banning certain deceptive election deepfakes even with a label, with penalties up to 5,000 dollars and five years. It also requires the state’s top election administrator to publicly correct credible falsehoods about voting.
Then notice the seam. Louisiana’s separate June 2026 ad-disclosure law excludes candidates for federal office. A fabricated video about a congressional candidate can go unlabeled in the state with the newest rule on the books.
The Video Side of This Is Its Own Story
Fake robocalls and fake candidate clips are two halves of one problem. We already traced what actually happened when a cloned video of a candidate went viral, and why a label barely slowed it down.

The gap between a law and a phone call
Three structural problems keep repeating:
- Enforcement is slower than a dialer. A robocall can reach 100,000 phones before breakfast. A fine arrives years later, if at all.
- A label does not stop circulation. In most states, a warning satisfies the disclosure rule, so the false claim keeps moving with a footnote attached.
- Attribution is hard. Spoofed caller ID and offshore routing made the New Hampshire call tough to trace. Voice cloning only lowers the barrier further.
There is a bigger fight in the background, too. In November 2025, 36 state attorneys general sent Congress a letter opposing the preemption of state AI laws. The robocall petition is a smaller version of the same question: when the federal government sets a floor, does it quietly become a ceiling?
Why a voice clone is more than noise
Robocalls usually get filed under “annoying.” Voice clones land somewhere else. A call in your own language, in a voice you recognize, telling you the wrong polling place or the wrong day is a direct attempt to remove your vote. The pattern is old: for decades, deceptive calls and flyers have targeted low-income, Black, immigrant, and other marginalized communities with false information about arrests, deportation, or the date of the election. What changed is the cost and the fluency.
The 2026 version is cheap, multilingual, and endless. The National Consumer Law Center also warns that AI-generated conversations can distort facts and steer people toward extreme content, leaning on the same hallucination problems that show up in every chatbot. A script that adapts to the person on the line is not a recording. It is a persuasion machine.
Whether the FCC grants the waiver is genuinely uncertain. The docket is open, the election is close, and the political incentives point in one direction.
What to watch before November 3
Three things to track:
- The FCC docket. Whether the agency grants the Club for Growth waiver, denies it, or simply sits on it through election day.
- State enforcement. Which attorneys general move first, and whether any 2026 case reaches a courtroom before the vote.
- Your own phone. If a political call arrives in a cloned voice, that is a data point that the guardrails failed.
If you get one, do not act on its instructions. Hang up, verify voting details with your state or county election office, and report the call to your state attorney general and the FCC.
FAQ
Are AI robocalls illegal right now?
AI-generated voices in robocalls are treated as “artificial” under the TCPA, so they are illegal without prior consent in many cases. The Club for Growth petition, if granted, would carve out political calls specifically.
Did the New Hampshire robocall change anything?
Partly. It produced a 6 million dollar FCC fine, criminal charges in New Hampshire, and a wave of state deepfake laws. It also showed how hard fines are to collect: Kramer said he would not pay the 22,500 dollar civil judgment.
Can a state ban a deepfake even if it is labeled?
Maryland, Minnesota, and Texas can prohibit certain deceptive deepfakes even when a warning is attached. In most of the 31 states with laws, a disclosure is enough and the content stays up.
How is this different from the fake candidate video story?
It is adjacent. Candidate clips circulate on social platforms where people can scroll past. Robocalls go straight to a phone, often to older or lower-turnout voters, and arrive with no feed to ignore. Same technology, different delivery.
The Short Version
- The FCC is taking comments on a Club for Growth petition to let AI-voice political robocalls reach cellphones without consent, four weeks before the midterms.
- The NCLC estimates the change could multiply political robocall volume by up to 40 times, toward 15 billion calls.
- The 2024 New Hampshire Biden clone cost about a dollar and under 20 minutes to make. It drew a 6 million dollar FCC fine and criminal charges.
- Federal law treats AI voices as “artificial,” and 31 states have election deepfake laws, but 28 of them accept a simple label.
- Enforcement is slow, collection is slower, and the preemption fight is still unresolved.

