A comprehensive and deeply concerning analysis warns that the already significant impact of artificial intelligence on political elections is poised to expand dramatically, transforming the very nature of democratic competition. Drawing on a series of noteworthy incidents from the 2025 off-year elections, technology policy experts and democracy advocates caution that these events represent merely a prelude to a more widespread and potentially destabilizing integration of AI into the 2026 midterm campaigns. The central thesis is that 2025 served as a crucial testing ground, demonstrating not only the technology’s capabilities but also the increasing willingness of political actors to deploy it. Consequently, experts characterize the developments of 2025 as just the “tip of the iceberg,” signaling an imminent future where AI-generated deepfakes and sophisticated misinformation campaigns become commonplace tools in the political arsenal, posing a formidable threat to the integrity of the democratic process.
Precedents from the Campaign Trail
The 2025 election cycle provided several stark examples that collectively illustrate the diverse and evolving applications of artificial intelligence in the political sphere, serving as concrete evidence for a broader trend of AI adoption. In the Virginia lieutenant governor’s race, one campaign took a novel approach to a common political challenge. After his Democratic opponent, state Senator Ghazala Hashmi, declined multiple debate invitations, Republican candidate John Reid utilized artificial intelligence to create a digital version of her. He then proceeded to hold a public debate against this AI-generated avatar of his rival. This event, while not overtly malicious, highlighted how AI can be used to reshape the very nature of political discourse and public engagement, creating new and unprecedented forms of political theater that blur the lines between reality and simulation. It showcased a strategic, non-combative use of the technology that nonetheless raised fundamental questions about representation and consent in political dialogue.
In a far more pernicious application of the technology, the New York City mayoral race witnessed a deeply troubling incident involving former Governor Andrew Cuomo. His campaign briefly published a deepfake advertisement that targeted his political opponent, Zohran Mamdani, who ultimately won the election. This ad employed the power of AI to generate a video laden with harmful and inflammatory racist stereotypes. Although the ad was quickly deleted, its brief existence served as a potent demonstration of how generative AI can be weaponized to execute character assassinations with a speed and potential virality that traditional attack ads cannot match. Perhaps most systemically dangerous, an incident in Utah saw AI used not to attack a candidate but to undermine the electoral process itself. The state’s Lieutenant Governor, Deirdre Henderson, was compelled to issue an urgent warning to voters after an AI-generated image depicting entirely fabricated election results began circulating online before the official polls had even closed. This represented a direct assault on voter confidence, illustrating how AI can be deployed to sow chaos and distrust in the foundational mechanics of democracy.
A Confluence of Technology and Timidity
Given these precedents from 2025, a profound sense of apprehension has solidified among experts regarding the 2026 midterm elections, which will feature a multitude of high-stakes gubernatorial and congressional races. The overarching consensus is that the political landscape in 2026 will be saturated with AI-generated deepfakes and meticulously crafted misinformation campaigns. Isabel Linzer, a policy analyst at the Center for Democracy and Technology (CDT), articulated this growing anxiety, asserting, “We’ve only seen the tip of the iceberg when it comes to AI’s impact on elections.” Linzer’s analysis points to a dangerous confluence of factors: the underlying AI technology is rapidly becoming more sophisticated and accessible, while simultaneously, both principled politicians and malicious actors are growing more comfortable and adept at using it. She specifically cautioned against a false sense of security that may have set in after the 2024 U.S. election, which managed to avoid any catastrophic AI-related incidents. This lack of a major crisis, she warns, should not be misinterpreted as an indication that the danger has passed; rather, it was merely a temporary reprieve before the technology’s full potential for disruption is realized.
State-level lawmakers have not been entirely idle in the face of this emerging threat, recognizing the potential for AI to corrupt political campaigns and deceive voters. According to research compiled by the National Conference of State Legislatures (NCSL), a significant number of states—26 in total—have already enacted laws specifically designed to regulate the use of political deepfakes. These new statutes generally fall into one of two categories: some impose outright bans on the creation or dissemination of deceptive AI-generated content in a political context, while others take a transparency-focused approach, mandating clear disclosure and labeling so that voters are aware when they are viewing synthetic media. Chelsea Canada, a program principal at NCSL specializing in technology, confirmed that regulating deepfakes will continue to be a “huge trend” within state legislatures. She noted that lawmakers are actively attempting to be proactive, adopting a targeted regulatory strategy to address the most immediate harms of AI while striving to protect the long-term integrity of their electoral systems. These legislative conversations are driven by constituent concerns and a desire to preemptively address the potential long-term, corrosive effects of unregulated AI on public trust.
The Crumbling Guardrails of Political Conduct
Despite these legislative efforts, recent research suggests the problem is far more complex and deeply rooted in evolving political norms. The CDT’s work highlights the multifaceted risks posed by generative AI, including its capacity to exponentially amplify disinformation, provide foreign adversaries with powerful tools for interference, and automate voter suppression campaigns on an unprecedented scale. While the technology also has positive applications for campaigns—such as assisting with data analysis or drafting communications—the potential for misuse presents a tremendous and possibly overwhelming risk to democratic health. Perhaps the most critical finding pertains to the relative restraint seen in the 2024 election cycle. The primary inhibitor on the harmful use of generative AI during that period was not legal or technological, but rather a set of self-imposed behavioral norms. According to Tim Harper, a co-author of the CDT report, campaign operatives largely refrained from using deepfakes because they held a strong belief that voters would recognize such tactics as unfair and would consequently penalize the offending candidates at the polls.
However, Harper delivers a sobering assessment: these crucial, self-policing norms are now actively “crumbling.” The evidence for this erosion is found in the increasingly common use of AI-generated content by some of the most prominent political entities in the nation. Harper points specifically to various AI-generated videos posted on social media by both the White House and former President Donald Trump. These videos, which often served to denigrate political opponents or protestors, have been condemned by some observers but have not generated the universal backlash that might have been expected in the past. This bipartisan adoption of AI-driven messaging, and the mixed public reaction to it, signals a dangerous shift. As high-profile political figures deploy these tools without suffering significant, career-damaging consequences, it effectively lowers the perceived political risk for all other campaigns. Harper explains that as these disincentives—the fear of voter punishment and widespread condemnation—continue to decline, the powerful incentives for campaigns to use every available advantage, including manipulative AI, become far more prominent and difficult to resist.
An Unsettling Path Forward
The existing framework of guardrails ultimately proved insufficient to contain the problem as it rapidly evolved. The calls for regulation alone could not serve as the sole solution, and the hope for campaigns to appeal to their own “better angels” was a difficult proposition in a highly competitive, high-stakes environment. With the unwritten rules that once encouraged restraint continuing to erode, and with many of the new state laws focused primarily on disclosure rather than prohibition, the deterrents were revealed to be weak. A simple “Made with AI” label did little to blunt the emotional impact of a viral deepfake designed to deceive or enrage. Furthermore, the major social media platforms and AI development companies faced weak political incentives to aggressively police their own systems, often fearing accusations of bias or censorship, which created a dangerous regulatory vacuum. The erosion of norms had become a bipartisan phenomenon that was set to escalate into 2026, creating a midterm election cycle that would be defined by an unprecedented onslaught of artificial intelligence in politics.
