It’s the homestretch of primary election season across the United States, and one of the key takeaways so far is that 2026 appears to be the year when political campaigns fully embrace the use of artificial intelligence. From the viral AI videos promoting Spencer Pratt’s campaign for mayor of Los Angeles to the use of deep-fake attack ads in the hotly contested U.S Senate race in Texas, AI-generated content has been abundant throughout the primaries and appears likely to accelerate as the November general election approaches.
Warnings about AI-generated misinformation upending American democracy have featured prominently in the lead up to the last few election cycles, and in response California, alongside 30 other states, enacted laws to regulate deceptive uses of AI in election communications. Yet the utilization of AI so far in 2026 suggests that much of the content circulating online remains beyond the reach of regulation. Even when content does fall within its purview, the capacity for the government to act in a meaningful way is severely limited.
Undeterred, congressional Democrats are already signaling their intention to pursue federal regulations on the use of AI in election advertising if they return to power in 2027. Before attempting another likely fruitless crusade, policymakers should pause and consider why state-level regulations have failed to stem the tide of AI content during recent primary elections.
The central and unavoidable hurdle to regulating the use of AI in elections is the fact that most state regulations of AI-generated election content impact political speech protected by the First Amendment. California was the first state to learn this lesson after Assembly Bill 2839 — the 2024 law banning the use of deceptive AI-generated election content within 120 days of an election — was struck down in federal court for violating the First Amendment.
While the government should respond to false claims about the election process — such as the promotion of incorrect polling locations or voter registration deadlines — it does not need to intervene in campaign disputes. Competition creates natural incentives for candidates and their supporters to respond to lies and misinformation, but there’s no such constituency to respond to false claims about election procedures. Yet, many of the state regulations are focused on the false AI-generated depictions of candidates, requiring enhanced scrutiny over their impacts to freedom of speech.
In an attempt to stay within the parameters of the First Amendment, most states have opted for regulations that do not prohibit speech but instead mandate disclosure when AI is used to create or modify election-related content in a deceptive manner. Yet even this approach is not guaranteed to withstand court scrutiny if the law does not include exceptions for things like satire and parody. Hawaii’s disclosure law, Act 191, was struck down in January for these very reasons.
Beyond the constitutional considerations, labeling requirements also face substantial implementation challenges that limit their effectiveness. For example, the viral AI-generated video depicting Spencer Pratt as Batman fell outside California’s regulatory framework because the video was created by a third party unaffiliated with the campaign. Even though it was unlabeled and Pratt himself promoted it online, the state’s disclosure requirement only applies to political advertisements subject to regulation by the California Fair Political Practices Commission (FPPC), which does not include a video created by a third party.
Another factor that hinders the effectiveness of disclosure requirements is that they require a court or government agency to determine whether a communication was AI-generated and assess the truthfulness of the underlying claim. These are simple concepts but difficult to implement. Pinpointing the technology used to generate or alter a video requires technical expertise, while assessing the accuracy of a statement about a candidate requires the government to investigate murky claims where the truth is not always obvious. Doing any of this requires resources, as evidenced by California FPPC’s budget request for $350,000 in ongoing funding to implement the state’s disclosure requirement, including $100,000 per year for a deep-fake detection software license.
As AI capabilities continue to advance and further integrate into the political process, the challenges states face in regulating AI in elections will only intensify. The same fundamental dynamics that limit the effectiveness of state regulations also apply at the federal level. Attempting to shield the public from exposure to false AI-generated claims using burdensome government regulations generates real costs with no discernible benefit.
A better course of action is to encourage free expression and debate while trusting that Americans will, once again, adapt to changes in an ever-evolving information landscape.
Chris McIsaac is a fellow with the R Street Institute’s governance program.