A major election in the U.S. is right around the corner and if you are like most people, you generally do not like all these campaign ads that seem to fill our screens. This cycle is different.  AI generated images, voices and videos have moved from novelty to campaign tools, from inexpensive local campaigns to the multimillion-dollar battles being fought in highly contested statewide contests.

Congress has not enacted a comprehensive federal law governing AI use in politics and it has left a void resulting in a patchwork of state-level efforts to fill the gap. Which means when it comes to AI in political ads, you have likely already seen it, but you might not have known if it was real or not.

In the past few months, campaigns in more than one state, in more than one race, from both parties, have run ads created by AI about their opponent. These can be misleading and deceptive. Misinformation in political ads is nothing new, but the use of AI is something politicians on both sides of the aisle have in many states recognized there should be a new set of rules. The problem is what those rules look like are inconsistent, right at the time they are needed most.

This is Different – Real Campaigns, Unreal AI

AI image generation has progressed steadily since the last election cycle. Early versions of generative AI pictures were a better version that photoshopped still image but many had flaws that made them easy to detect. In 2024 AI generated video was nearly impossible with high quality. Tools like Google’s Veo and OpenAI’s Sora barely existed in 2024 and were introduced in May and December of that year. Those early systems struggled with consistency, duration, and realistic movement. Two years later, the technology creates high-quality images, including sound and cloned voices that make synthetic media difficult to distinguish from authentic video, at a faction of the cost and effort. Where we once trusted our instincts of what we could see and hear, we now need to question what is real.

“…Tools like Google’s Veo and OpenAI’s Sora barely existed in 2024…”

On the surface this should be simple. Either a state bans AI entirely or allows it with sufficient AI labelling and disclosure. Thirty-one states have now enacted laws regulating deepfakes in political messaging, but they have taken very different approaches. Most have opted for permitted use and disclosure. About a third ban AI in some way. Texas, for example, prohibits certain deceptive deepfake videos distributed only within 30 days of an election. Beyond this period AI videos are permissible. Other states use different time windows or require disclosures rather than outright prohibitions.

In Montana a federal judge recently blocked enforcement of the state’s AI law against a political group that challenged it, finding that is did not align with the First Amendment. The ruling did not change or invalidate the law statewide. In effect a case-by-case application of First Amendment rights, AI, and politics.

The issue becomes more layered. AI laws in two states, California and Hawaii, have been struck down, and others in Minnesota and Montana are currently being challenged and could face the same fate. States such as Texas are also challenging how the law is interpreted with regard to the 30-day ban period and if that infringes on early-voting. This can be confounding for voters and candidates alike.

Intent and Truth

Today, neither the FTC nor Congress has taken up the issue of AI advertising in political campaigns. The FTC is the federal agency that has domain over truth in advertising – they’re the reason when Tide says it cleans clothes 4 times better, Tide needs to prove it. However, the First Amendment provided broad latitude, making regulation more difficult and gives politicians an exit ramp on verified truth claims in their ads.

This is not new and it’s a reason that since TV political ads debut in 1952, people have had a strained relationship with political advertising. While voters want to know more about candidates, a national survey published in 2024 rated campaign advertising nearly 6 out of 7 for being annoying, excessive, and misleading. Misinformation couldn’t have a more welcome home.

Many of the disputes of the current laws center around a common exception – intent. At least nine states allow exceptions to their AI laws that permit its use when the imagery is intended to be parody or satire. Arizona law for example, requires proof that the AI creator knew of its falsity, intended to injure the reputation of the candidate, and that it intentionally misleads a reasonable person. AI that is obviously cartoonish or humorous gets a pass, but what happens when those intents are not obvious, when AI appears so real, a reasonable person cannot tell the difference?

Beyond Boundaries

Most people claim that political ads have no influence on the way they vote. I have written before how AI can change your opinion on topics because of its ability to customize messages and its human-like-conversational-trust mechanism.

Seeing makes it easier to believe. An Ipsos study unrelated to politics, found that only 13% of people could confidently identify an AI created ad. The same percentage thought a real ad was actually created by AI. AI confusion and authenticity have reached a new level since the last election cycle. That confusion can still be effective even when we know they might not be real because visualization can confirm our bias and shape our decisions. Those ads are no more informing than distracting and the various state laws that seek to contain them neatly within their geographical borders will be no match for the next election that goes beyond those boundaries.