AI deepfake ads grow more popular in US midterm campaigns, blurring truth

The US midterm election campaign has been filled with political ads featuring AI deep fakes that can mislead voters.

US political candidates and their funders have spent more than $80m to make AI-based ads this election season, with over 80 percent coming from Republicans [File: Jonathan Drake/Reuters]

San Francisco, USA: Bruce Blakeman, a Republican who is running for governor of New York, released campaign ads featuring Zohran Mamdani, New York City mayor, and Kathy Hochul, the state’s governor, gardening together and collaborating to raise prices for residents.

Another ad features Hochul talking about hiking tolls and fares so people cannot afford to drive or take the subway.

list 1 of 4 Who is Navi Pillay, winner of the 2026 Nobel Peace Prize?

list 2 of 4 US to livestream Fort Hood shooter’s execution: Is that legal?​

list 3 of 4 ‘Rogue court’: US sanctions ICC hours after ex-judge wins Nobel Prize

list 4 of 4 Trump launches probe into Federal Reserve Governor Lisa Cook

Neither Hochul nor Mamdani was actually in the ads, which were made using artificial intelligence (AI). A spate of AI-generated ads that are obviously fake but grab voter attention have been a feature of campaigns in the lead up to the pivotal United States midterm elections on November 3.

While regulation has focused on labelling AI deepfakes that could mislead voters, it is “slopaganda” – or identifiably fake ads – that mark this campaign season, says Scott Babwah Brennen, director of New York University’s Center on Technology Policy and co-author of a recent report, Political Slopaganda.

Political candidates and their supporters have spent more than $80m to make more than 164 AI-based ads this election season, according to the Wesleyan Media Project, a non-partisan election advertising tracker. Republican candidates and their backers were behind 80 percent of these ads, the group found.

Blakeman defended his ads, saying they were clearly labelled as AI-generated – a disclaimer the ads do carry for a few seconds at the start – and follow a long tradition of satire and parody in American politics.

But slopaganda – a mix of low-quality AI “slop” and propaganda – is being used to “communicate deeper truths rather than factual truths”, said Brennan, meaning it could shape perceptions of a candidate’s character or ideology and could “impact voter trust”.

Other such ads showed Sherrod Brown, a Senate Democratic candidate from the state of Ohio, as the last guest at a party, having overstayed his welcome.

James Talarico, a Senate Democratic candidate from the state of Texas, was shown in a blue dress similar to the character Maria from the movie The Sound of Music, while Mike Rogers, a Senate Republican candidate from Michigan, appeared as a superhero.

In San Francisco, when Democratic California state Senator Scott Wiener released an AI chatbot in the high-profile race to succeed House Speaker Emeritus Nancy Pelosi, who represents the tech-industry city in Congress, it drew an immediate backlash.

The chatbot answered around 5,000 questions about his opponent, Connie Chan, parodying her record of ostensibly blocking real estate development in the city, her accent as well as her naturalized citizenship. The move sparked criticism for being bigoted and Wiener withdrew the chatbot, saying that while it was innovative, “it missed the mark”.

Deepfake ads have emerged in a landscape that aims to protect political speech, including when done through parody. While there is no federal law regulating ads made with AI deepfakes, 31 states have passed their own laws calling for ads made using AI to be labelled.

“State laws do what they can without encroaching on the First Amendment,” said Ilana Beller, director of State AI Governance and Policy at Public Citizen, a Washington, DC, based consumer rights advocacy group. “There is still a need for a federal law,” she said.

Adam Rose, chief operating officer of Starling Lab, a research centre on technology and human rights co-founded by Stanford University and the University of Southern California, said some changes made by AI could just be colour or voice correction that the government does not need to label. AI-generated ads should not be banned, he said.

Instead, Rose said, candidates are always involved in making “judgement calls” on the ads and these calls reflect on them. For Wiener, Rose said, releasing the chatbot “showed bad judgement. It showed voters something about him”.

Wiener is running to represent San Francisco in the US Congress, a city that is the global epicentre of the AI industry with two of the biggest AI companies, Open AI and Anthropic, based there. The city’s representative will play an important role as legislation around AI emerges, especially as polls show a majority of Americans are more concerned than excited about the technology.

“This is a transition moment as AI is reshaping our society,” says Ian Krager, a spokesperson for Connie Chan. “This race is a harbinger for the country. It is a moment where we can show the world America can lead the world on AI development without asking working people to bear the cost.

“And it is clear Connie Chan is with working families while Scott Wiener is with the AI companies,” he said, citing the AI chatbot.

As a state senator, Wiener drafted a California law last year requiring AI companies to disclose security incidents where autonomous systems created by extremely large AI models are involved in dangerous or harmful incidents. A less stringent version became law last year. His office did not respond to Al Jazeera’s request for comment.

Labels as clear as food labels

Looking at the final stretch of the November midterms and beyond, Public Citizen’s Beller said there is a need to make labelling of ads that use AI as clear and easily understood as food labels so voters can evaluate them accordingly.

“People need to know to look for labels,” she said, as it’s important to create awareness about AI deepfake ads, especially among younger voters and seniors.

In New York, the Democratic Party has filed a complaint with the state board of elections saying the ads violate disclosure requirements and are “misleading to the voters”. But Blakeman has said that “people find it humorous. They like satire”.

The ads are “trying to get attention when candidates know there is political power to be gained by getting attention”, said NYU’s Brennen.

With AI deepfake ads only growing, some candidates have publicly said they will not use them. “People seem to like [the candidates] for it,” said Beller.

美国中期选举竞选期间充斥着大量利用人工智能深度伪造技术制作的政治广告，这些广告可能会误导选民。

美国政治候选人及其资助者在本届选举季已花费超过 8000 万美元用于制作人工智能广告，其中超过 80% 来自共和党人。[图片：Jonathan Drake/路透社]

美国旧金山：竞选纽约州州长的共和党人布鲁斯·布莱克曼发布了竞选广告，广告中纽约市市长佐兰·马姆达尼和纽约州州长凯西·霍楚一起种花，并合作提高居民的物价。

另一则广告中，浩哲谈到要提高过路费和票价，让人们负担不起开车或乘坐地铁的费用。

列表 1/4：2026 年诺贝尔和平奖得主纳维·皮莱是谁？

美国四所大学中的第二所将直播胡德堡枪击案凶手的死刑执行：这合法吗？

列入“流氓法庭”名单的3/4：美国在国际刑事法院前法官获得诺贝尔奖数小时后对其进行制裁

特朗普对美联储理事丽莎·库克展开调查（共4条，第4条）

霍楚和马姆达尼实际上都没有出现在这些广告中，这些广告是由人工智能（AI）制作的。在11月3日关键的美国中期选举前夕，大量明显造假但却能吸引选民眼球的AI生成广告已成为竞选活动的一大特色。

纽约大学技术政策中心主任、近期报告《政治宣传》的合著者斯科特·巴布瓦·布伦南表示，虽然监管的重点是给可能误导选民的人工智能深度伪造视频贴上标签，但真正标志着本竞选季的是“虚假宣传”（或可识别的虚假广告）。

据无党派选举广告追踪机构卫斯理安媒体项目（Wesleyan Media Project）的数据显示，本届选举季，政治候选人及其支持者已花费超过8000万美元制作了164条基于人工智能的广告。该机构发现，其中80%的广告是由共和党候选人及其支持者投放的。

布莱克曼为自己的广告辩护说，这些广告明确标明是人工智能生成的——广告开头确实有几秒钟的免责声明——并且遵循了美国政治中讽刺和模仿的悠久传统。

但布伦南表示，低质量人工智能“垃圾”与宣传相结合的“垃圾宣传”正被用来“传达更深层次的真理，而不是事实真相”，这意味着它可以塑造人们对候选人性格或意识形态的看法，并可能“影响选民的信任”。

其他类似的广告则将俄亥俄州民主党参议员候选人谢罗德·布朗描绘成派对上最后一位客人，因为他待得太久而无人欢迎。

来自德克萨斯州的民主党参议员候选人詹姆斯·塔拉里科身穿一件类似电影《音乐之声》中玛丽亚角色的蓝色连衣裙亮相，而来自密歇根州的共和党参议员候选人迈克·罗杰斯则装扮成超级英雄。

在旧金山，当民主党加州参议员斯科特·维纳 (Scott Wiener) 在备受瞩目的众议院议长竞选中推出人工智能聊天机器人时，立即引发了强烈反对。维纳是代表这座科技产业城市的国会议员，而众议院议长是前众议院议长南希·佩洛西 (Nancy Pelosi)。

这款聊天机器人回答了大约5000个关于其对手陈康妮的问题，讽刺了她阻挠该市房地产开发的记录、口音以及她的入籍身份。此举引发了关于其带有偏见的批评，维纳随后撤回了这款聊天机器人，并表示虽然它很有创意，但“却偏离了主题”。

在旨在保护政治言论（包括戏仿言论）的背景下，深度伪造广告应运而生。虽然目前尚无联邦法律监管使用人工智能制作的深度伪造广告，但已有31个州通过了各自的法律，要求对使用人工智能制作的广告进行标注。

“各州法律在不侵犯宪法第一修正案的前提下尽力而为，”总部位于华盛顿特区的消费者权益倡导组织“公众公民”（Public Citizen）的州人工智能治理与政策主任伊拉娜·贝勒（Ilana Beller）表示，“但我们仍然需要一部联邦法律。”

斯坦福大学和南加州大学联合创办的科技与人权研究中心Starling Lab的首席运营官亚当·罗斯表示，人工智能所做的一些修改可能只是颜色或语音校正，政府无需对此进行标注。他认为，不应禁止人工智能生成的广告。

罗斯表示，候选人始终参与广告的“判断”，而这些判断会反映出他们的立场。罗斯认为，维纳推出聊天机器人“表明他判断失误，这让选民对他产生了误解”。

维纳正在竞选代表旧金山进入美国国会。旧金山是全球人工智能产业的中心，两家最大的人工智能公司——OpenAI和Anthropic——都总部设在这里。随着人工智能相关立法的不断涌现，旧金山的代表将扮演重要角色，尤其是在民调显示大多数美国人对这项技术更多的是担忧而非兴奋的情况下。

“人工智能正在重塑我们的社会，这是一个转型时刻，”陈康妮的发言人伊恩·克雷格说。“这场竞赛对国家来说意义重大。这是一个我们可以向世界展示美国能够在人工智能发展方面引领世界，而无需让劳动人民承担成本的时刻。”

“很明显，康妮·陈 (Connie Chan) 代表的是工薪家庭，而斯科特·维纳 (Scott Wiener) 则代表的是人工智能公司，”他说道，并提到了人工智能聊天机器人。

作为加州州参议员，维纳去年起草了一项加州法律，要求人工智能公司披露由超大型人工智能模型创建的自主系统卷入的危险或有害事件中的安全事故。去年，一项限制较少的版本正式生效。他的办公室没有回应半岛电视台的置评请求。

标签清晰明了，如同食品标签一般。

展望 11 月中期选举的最后阶段以及未来，公共公民组织的贝勒表示，有必要让使用人工智能的广告标签像食品标签一样清晰易懂，以便选民能够据此进行评估。

“人们需要知道如何查看标签，”她说，因为提高人们对人工智能深度伪造广告的认识非常重要，尤其是在年轻选民和老年人中。

在纽约，民主党已向州选举委员会提出申诉，称这些广告违反了信息披露规定，并且“误导选民”。但布莱克曼表示，“人们觉得这很有趣。他们喜欢讽刺”。

纽约大学的布伦南表示，这些广告“试图引起关注，因为候选人知道，引起关注可以获得政治权力”。

随着人工智能深度伪造广告的日益普及，一些候选人已公开表示不会使用此类广告。“人们似乎更喜欢（这些候选人）这样做，”贝勒说道。