AI deepfakes and robocalls test election rules in 2024

AI stopped being an abstract workplace story in 2024 and became a live test of democratic resilience. The year exposed how synthetic voices, fabricated images, and automated influence can be used not just to confuse voters, but to suppress turnout, shift campaign strategy, and weaken trust in the system itself.

2024 became the proving ground

The most important lesson from the year is that AI’s political threat is broader than job losses or productivity gains. Election officials, lawmakers, platforms, and watchdogs confronted a new reality in which deceptive content could be created quickly, cheaply, and at scale, then pushed into the information stream before anyone could verify it.

That is why 2024 mattered as a proof point. In one case, AI was used to impersonate a sitting president. In another, a deepfake ad reportedly landed in a down-ballot race and altered a low-profile campaign’s strategy. Together, those episodes showed that synthetic media was no longer a theoretical risk. It had become a practical campaign tool.

New Hampshire showed how turnout could be targeted

The clearest early warning came in New Hampshire. On January 21, 2024, voters received robocalls that used what appeared to be an AI-generated voice clone of President Joe Biden urging them not to vote in the January 23 Democratic presidential primary. The New Hampshire Department of Justice identified the source of the calls, and Attorney General John Formella’s office treated the episode as a serious election interference case.

The Federal Communications Commission later fined political consultant Steve Kramer $6 million in September 2024 for the illegal robocalls and caller-ID spoofing. He was also facing criminal charges in New Hampshire. That sequence matters because it shows the full chain of the problem: synthetic voice creation, deceptive delivery, and an enforcement response that arrived only after the calls had already circulated.

The episode was more than a one-off stunt. It showed that AI can be used to suppress participation, not just spread confusion. A fake message telling people not to vote cuts at the core of democratic process, and it can do so before campaign staff, journalists, or election officials have time to react.

Washington chose existing rules over a new AI regime

At the federal level, regulators moved cautiously. On September 19, 2024, the Federal Election Commission said it would not open a new rulemaking on AI in campaign ads. Instead, it adopted an interpretive rule saying existing anti-fraud regulations already apply to deceptive AI uses in campaign advertising.

That decision reflects a broader tension in election policy. Lawmakers and regulators can either build new rules tailored to AI or argue that existing laws are flexible enough to catch the abuse. The FEC chose the second route, which may have reduced immediate regulatory friction, but it also left unresolved how aggressively new synthetic tactics should be policed as the technology evolves.

The practical question is not only whether deceptive content is illegal. It is whether the rules are clear enough, fast enough, and visible enough to deter bad actors before the damage is done. In 2024, the answer was only partly yes.

States moved faster than the federal government

State governments proved more active. The National Conference of State Legislatures said at least 45 states, Puerto Rico, the Virgin Islands, and Washington, D.C. introduced AI bills in the 2024 legislative session. It also said 31 states plus Puerto Rico and the Virgin Islands adopted resolutions or enacted legislation.

California was the most prominent example of a state trying to get ahead of the threat. On September 17, 2024, Governor Gavin Newsom signed three bills aimed at deceptive AI election content. According to the governor’s office, the measures were meant to remove deceptive content from large online platforms, increase accountability, and better inform voters.

But even in California, the legal path was not straightforward. One of the new laws, AB 2839, later faced a federal injunction challenge over First Amendment concerns. That challenge underscores a central problem in election law: guarding against manipulation without sliding into overbroad speech restrictions. The result is a policy race in which states are experimenting, but courts still have a final say on how far they can go.

Platforms and AI companies built guardrails, but not perfect ones

Technology companies also tried to answer the problem from inside the system. OpenAI said in January 2024 that it was approaching worldwide elections by working to prevent abuse, provide transparency around AI-generated content, and improve access to accurate voting information. Later, it said it disrupted more than 20 operations and deceptive networks that attempted to use its models in 2024.

Meta also said political ads using AI-generated imagery on Facebook and Instagram would have to disclose that fact. That is a meaningful step because disclosure can give voters at least a first layer of context, especially when the content looks plausible enough to pass casually through a feed.

Still, guardrails are only as strong as enforcement and transparency. Platforms can label, remove, or demote content, but they cannot fully control how quickly a deepfake spreads once it is copied, reposted, or repackaged elsewhere. A disclosure rule helps, but it does not eliminate the persuasive power of a polished falsehood.

The real damage is still hard to measure

Researchers have been blunt about the limits of the evidence. AI election harms are difficult to measure because data gaps are large and platform transparency remains limited. That makes the political impact harder to quantify than the technical capability. It is often easier to identify the fake than to calculate how many minds it changed.

That uncertainty matters for policy. If lawmakers cannot measure the scale of the problem, they can underestimate it. If they overreact to dramatic edge cases, they risk writing blunt rules that miss the subtler ways AI can shape campaign spending, persuasion, turnout, and voter trust.

The deeper risk is not only that voters will see falsehoods. It is that they will stop believing what is real. Once candidates can be impersonated, images can be fabricated, and turnout messages can be forged, every viral claim becomes slightly harder to trust. That is a democratic cost that outlasts any single election.

What the next cycle is likely to demand

The next national election cycle will test whether the guardrails built in 2024 can keep up with faster models, cheaper generation tools, and more sophisticated distribution tactics. The FEC’s reliance on existing anti-fraud rules, California’s attempt to legislate disclosure and platform accountability, and platform labeling policies all point in the same direction: the system is trying to adapt in real time.

But 2024 also made clear what still fails. Enforcement often arrives late. Disclosure rules do not always reach the most viral copies. And even successful takedowns do not repair the broader erosion of trust once voters have been exposed to a convincing fake.

That is the lasting lesson from the year’s deepfakes and robocalls. AI has already moved into the mechanics of campaigning, and the U.S. will enter the next election cycle with stronger instincts than it had in January 2024, but not with full control. The democratic stress test has begun, and the harder work is still ahead.