AI blurs lines in campaign ads: ‘People are seeing things that didn’t happen’

Dramatic political attack ads hitting the airwaves this year have shown a Senate candidate driving a bus of undocumented immigrants to a polling place, a governor hopeful grabbing bags stuffed full of cash, and a Republican congressman holding hands with two of his Democratic colleagues in what a narrator described as a “throuple.”

None of those compromising situations ever happened — the ads were all generated with artificial intelligence.

From January through mid-September, political campaigns and groups had spent more than $16 million on TV ads that disclosed using AI-generated or manipulated images, according to a CNN analysis of ad data.

But only about half of states require campaigns to disclose AI use in political ads — and at least $38 million has been spent on ads that strongly appear to include AI imagery despite featuring no disclosure.

That includes ads in key midterm races that could sway the balance of power in Congress, such as ads from GOP campaigns and groups showing Georgia Sen. Jon Ossoff getting a makeover and Texas Senate candidate James Talarico defacing a Bible . The Ossoff ad didn’t have any disclosure about AI use, while the Talarico ad featured a small disclosure saying that “certain video content” was created with AI.

The flood of ads containing lifelike imagery has sparked alarms among lawmakers and voters alike, fearful of being misrepresented and misinformed. Already, the new frontier of AI ads has roiled political campaigns, with some ads being taken off air after complaints by candidates, and a Texas candidate filing a police report alleging that a video depicting her endorsing a list of new taxes violated state law.

AI’s use in campaign commercials comes amid public unease with the underlying technology and growing debate among policymakers over how, and whether, to take additional steps to rein it in. A new CNN poll found that about eight in 10 Americans express negative views about the rapid pace of AI development.

CNN’s analysis found that Republicans have been quicker to adopt AI in political ads than their Democratic opponents. About 96% of the spending on ads that disclosed AI use came from GOP campaigns or groups.

CNN found a handful of ads that appear to skirt state laws about disclosures. In Michigan, a PAC linked to Texas Sen. Ted Cruz, a Republican, aired an ad showing Senate hopeful Abdul El-Sayed laughing at messages saying “Death to America” and “Celebrate 9/11,” and dancing with the left-wing influencer Hasan Piker and other figures.

The final seconds of the ad included a disclosure in tiny text and a spoken line acknowledging it was “created in whole or in part with the use of artificial intelligence” — but one Michigan law requires “materially deceptive” ads that aim to influence elections to feature a disclosure “throughout the entirety of the video” that is “clearly visible to and readable by an observer.”

In a statement to CNN, El-Sayed campaign spokesman Cole Wozniak said Republicans “are so desperate that they are resorting to cheap AI generated ads filled with falsehoods and bad faith attacks.”

The use of AI in political ads ranges from relatively minor changes like animating old family photos to surreal attacks, like one ad that put a Florida governor candidate’s talking head on the body of a fish or another that showed President Donald Trump frying a patty with a congressman’s face on it.

While some of the most outlandish ads are obviously AI-generated, experts are worried about less clear cases.

The danger is that “people are seeing things that didn’t happen, seeing politicians say things they actually didn’t say,” said Travis Ridout, a political scientist at Washington State University and one of the co-directors of the Wesleyan Media Project, which has studied AI use in political ads.

“Even if there’s a disclaimer on there that says this was synthetically generated, oftentimes seeing is believing,” he said.

To track the growth in spending in AI political ads, CNN analyzed thousands of ads that aired this year and were collected by the tracking service AdImpact. Reporters used AI to extract all text shown on screen in each ad and then manually reviewed those with relevant disclosures. The analysis revealed dozens of ads that disclosed to viewers that they included AI-generated, enhanced or manipulated content.

Reporters also reviewed and identified several dozen additional ads that did not disclose using AI but strongly appear to feature AI-generated imagery. Some of those ads were flagged by researchers at the Wesleyan Media Project . Reporters excluded some ads when it was unclear whether they used AI or more traditional graphic effects technology.

The analysis found that spending on broadcast TV political ads that clearly featured AI grew from less than $25,000 in January to more than $10 million in the first half of September alone.

It’s likely that CNN’s analysis missed some ads that used AI without disclosing it, because the use of the technology is not always obvious.

A leading Republican super PAC, Senate Leadership Fund, produced an attack ad last month targeting Ossoff that uses AI imagery to mock his campaign spending on hair and makeup.

“Thousands on hair and makeup. What a phony,” claims an individual in the ad, which depicts Ossoff receiving cosmetic treatments including a blow dry, eye cucumbers and nail filing.

The Georgia Democrat called for the ad to be taken down at a press conference shortly after it was launched, saying it was “deeply troubling to see these deepfake videos used to fabricate political advertising.” An SLF spokesman said it was Ossoff’s record that was “deeply troubling.”

Another ad launched last month by Texas Attorney General Ken Paxton shows Talarico, his Democratic opponent for US Senate, drawing in a Bible and making statements about religion and gender.

The ad features quotes that Talarico said over the years without context — yet they’re spoken not by the actual Talarico, but an AI-generated version of him.

“Ken Paxton and his billionaire backers in DC know they can’t rely on facts to win this race — so their only answer is to literally put words in James’ mouth with AI deepfake videos to try and fool Texans,” Talarico spokesperson JT Ennis said in a statement.

Paxton’s campaign rejected the criticism.

“Our team was diligent in ensuring every piece of the ad was properly sourced and cited,” Paxton spokesperson Madison Cercy said in a statement. “We intend to continue to use all available means to put James Talarico’s radical record in front of Texans.”

AI advertising was also a flashpoint in the contentious race to oust GOP Rep. Thomas Massie in Kentucky’s 4th Congressional District earlier this year.

One of the outside groups opposing Massie ran ads that featured AI-generated imagery of the congressman walking hand-in-hand with Reps. Alexandria Ocasio-Cortez and Ilhan Omar, progressive Democrats vilified by conservatives, but who aligned with Massie on opposition to US military aid for Israel.

The ad showed what appeared to be paparazzi photos of the trio dining together, and security footage of them checking into a hotel, as a narrator declared that Massie was “cheating with the squad on the America First movement.” A small text disclosure shown during the ad’s first seconds stated that “this satirical ad was created with artificial intelligence.”

After he lost his primary earlier this year, Massie suggested the ad could have contributed to his defeat. “Many older people voted against me because they believed I dated AOC and Omar,” the congressman wrote about the ad in a post on X. “I told AOC I would never do that with Omar!” he joked.

Beyond the more contentious examples, many candidates have also turned to AI tools to simply and cheaply produce less controversial campaign content. Rick Jackson, the GOP nominee for Georgia governor, has spent millions this year on ads that include AI animations of family photos and dramatizations of his hardscrabble childhood, making his campaign the top spender on AI ads identified by CNN. (While some of his ads have disclosed using AI, others have not — disclosures are not required in Georgia.)

Nathaniel Persily, a Stanford law professor who studies the use of AI in elections, said that deepfake AI ads could end up having the most impact not in prominent statewide races but down-ballot, low-information contests where voters might not take the time to study candidates or fact-check ads they see.

“We’ve had Photoshop and attempts to manipulate imagery and lies that politicians have told since the beginning of time,” Persily said. “There’s something about realistic video that makes it more believable to people and has a greater emotional appeal.”

At least 31 states have passed laws regulating the use of AI content and deepfakes in political advertising, ranging from outright prohibitions to disclosure requirements, according to a tally by the National Conference of State Legislatures. Most of those laws were signed in the last three years, as AI use became more prevalent.

Most of the states require some sort of disclosure of AI content or deepfakes in political ads, while three states — Maryland, Minnesota and Texas — have banned some deepfake ads. Minnesota’s and Texas’ prohibitions only apply in specific periods leading up to elections.

“This year is a test case to see if those laws are being effective, are they being enforced, whether AI really becomes a major factor in elections,” said Katy Owens Hubler, the director of the NCSL’s elections program.

At least a few of the AI-powered ads that have aired in midterm races in recent months appear to violate those laws.

In Tennessee, several campaigns have been hit with AI attack ads, even after the state passed a law requiring disclosure in deepfake political ads in July.

Just weeks after that law passed, a dark money group opposing GOP governor candidate Marsha Blackburn aired an ad using AI without disclosure showing her cavorting with pharma lobbyists, tarring her as “Queen Mama Pharma” and tying her to opioid deaths in the state. Blackburn’s campaign sent a cease-and-desist letter to the group calling the ad false and defamatory and successfully got several TV stations to take the ad off the air.

In the state’s 6th Congressional District, two competing Republican candidates also attacked each other with dueling AI-powered ads. Johnny Garrett aired an ad showing someone with the nametag of his rival, Van Hilleary, receiving a stack of money from Michael Bloomberg, while Hilleary’s campaign hit back with an ad showing Garrett sleeping in office and attending a pride parade.

Both campaigns told local media that they believed the new deepfake law didn’t apply to candidates for federal office.

In Michigan, meanwhile, El-Sayed’s campaign has sharply criticized the AI-made ad, which seeks to closely link the candidate to the Muslim Brotherhood and Piker, a left-wing Twitch streamer.

A spokesperson for The Front Line Action, the super PAC behind the ad, defended the spot. “The substance of the ad is grounded in El-Sayed’s own public statements and documented reporting and records concerning him and individuals associated with or supportive of his campaign,” the PAC spokesperson said in a statement.

Also disputed: Which statute applies to the ad.

One Michigan law requires a three-second voiceover acknowledging AI’s use in political commercials and a four-second written disclosure. A second election law , governing “materially deceptive media,” however, also requires disclosure throughout a video if it “falsely represents a depicted individual,” intends to harm a candidate’s electoral chances and is distributed within 90 days of an election. Additionally, the person distributing the ad has to know that it falsely represents the person depicted. A PAC spokesperson said the group views the first law as governing its ad and said the commercial is in compliance.

Efforts to prohibit the distribution of deceptive AI-generated ads or mandate disclosures of AI use in advertising have failed to gain traction in Congress or with the Federal Election Commission, which currently lacks a quorum to operate anyway.

And legal experts say there are high hurdles to regulating political speech, particularly when ads reflect words that the targeted candidates actually uttered.

Generally, AI-generated ads that can be interpreted as satire or parody are exempt from the state disclosure requirements, said Ilana Beller, director of state AI governance and technology at Public Citizen, the left-leaning consumer advocacy group. The organization has advocated for regulation at the federal and state level .

Federal judges in Hawaii and California found parts of election deepfake laws in those states unconstitutional on First Amendment grounds.

But even without federal action, the growing number of state regulations seems to have helped limit what Beller called the “nightmare scenario” of truly deceptive, realistic-looking commercials in which “the public sees a thing and believes that the candidate actively participated in something morally wrong or scandalous in some way that would sway public opinion.”

Controversies aside, AI has had the broad effect of lowering the barrier to entry to political ad-making.

Political advertisers and observers said that ad-makers have been turning to AI to streamline their process, saving money and time by testing AI versions of ads that, if approved, go into real production.

Lauren French, the communications director of Senate Majority PAC, a leading Democratic group, said that while some political campaigns have always used ads to promote lies, they’re now “doing it with tools that were previously not available.” She criticized the ads attacking Talarico and Ossoff as misleading voters.

“We still need to embrace it and use it,” she said of the new technology. “But we can use it and be ethical, tell the truth.”

Meanwhile, other campaigns airing ads featuring genuine videos of their opponents have taken the time to explicitly declare that they aren’t generated by AI — in an attempt to help viewers cut through the slop.

“I’m going to show you something, no AI, no special effects, just Michael Whatley in his own words,” Democratic Senate candidate Roy Cooper declared at the beginning of one attack ad aired earlier this year.

Persily, the Stanford professor, said voters could eventually get so used to AI attacks that they would grow to doubt even real videos.

“It is going to be increasingly difficult for voters to know whether something is true or not,” he said. “That’s the greater harm to democracy.”

CNN’s Jay Brown contributed to this report.

今年，一些引人注目的政治攻击广告出现在电视上，广告中包括：一位参议院候选人驾驶一辆载有非法移民的巴士前往投票站；一位州长候选人抓起装满现金的袋子；以及一位共和党国会议员与他的两位民主党同事手牵手，旁白将其描述为“三人行”。

这些令人尴尬的情况从未发生过——所有广告都是由人工智能生成的。

根据 CNN 对广告数据的分析，从 1 月到 9 月中旬，政治竞选活动和团体在电视广告上花费了超过 1600 万美元，这些广告披露使用了人工智能生成或篡改的图像。

但只有大约一半的州要求竞选活动披露在政治广告中使用人工智能的情况——而且至少有 3800 万美元被用于那些明显包含人工智能图像但没有任何披露的广告上。

这其中包括一些可能左右国会权力平衡的关键中期选举广告，例如共和党竞选团队和团体发布的广告，其中一则广告展示了佐治亚州参议员乔恩·奥索夫的形象改造，另一则广告则展示了德克萨斯州参议员候选人詹姆斯·塔拉里科损毁圣经的画面。奥索夫的广告没有披露人工智能的使用情况，而塔拉里科的广告则包含一条简短的披露信息，称“某些视频内容”是由人工智能生成的。

大量逼真的人工智能广告引发了立法者和选民的担忧，他们担心自己被误导和歪曲事实。人工智能广告这一新兴领域已经搅乱了政治竞选，一些广告在遭到候选人投诉后被撤下，一位德克萨斯州候选人甚至向警方报案，称一段她支持一系列新税种的视频违反了州法律。

人工智能在竞选广告中的应用正值公众对这项技术本身感到不安之际，政策制定者之间也就如何以及是否采取更多措施来限制人工智能展开了日益激烈的辩论。美国有线电视新闻网（CNN）的一项最新民调显示，约有八成美国人对人工智能的快速发展持负面看法。

CNN的分析发现，共和党人在政治广告中采用人工智能的速度比民主党人更快。在披露使用人工智能的广告支出中，约96%来自共和党竞选团队或团体。

CNN发现一些广告似乎规避了各州关于信息披露的法律。在密歇根州，一个与德克萨斯州共和党参议员特德·克鲁兹有关联的政治行动委员会播放了一则广告，广告中参议员候选人阿卜杜勒·埃尔-赛义德对着写有“美国去死”和“庆祝9/11”等标语的画面大笑，并与左翼网红哈桑·皮克尔等人共舞。

广告的最后几秒包含一个很小的文字披露和一句口白，承认该广告“全部或部分是使用人工智能创作的”——但密歇根州的一项法律要求，旨在影响选举的“实质性欺骗性”广告必须在“整个视频中”包含一个“清晰可见且易于观察者阅读”的披露信息。

埃尔-赛义德竞选发言人科尔·沃兹尼亚克在给 CNN 的一份声明中表示，共和党人“已经走投无路，只能诉诸于充斥着谎言和恶意攻击的廉价人工智能生成的广告”。

人工智能在政治广告中的应用范围很广，从相对较小的改动（例如为旧家庭照片添加动画）到超现实的攻击，例如一则广告将佛罗里达州州长候选人的讲话头像放在鱼的身上，或者另一则广告显示唐纳德·特朗普总统正在煎一个印有国会议员头像的肉饼。

虽然一些最离谱的广告显然是人工智能生成的，但专家们担心一些不太明显的案例。

华盛顿州立大学政治学家、卫斯理媒体项目联合主任特拉维斯·里杜特表示，危险在于“人们会看到一些从未发生过的事情，看到一些政客说了一些他们实际上从未说过的话”。该项目研究了人工智能在政治广告中的应用。

“即使上面有免责声明说这是合成生成的，但眼见为实，”他说。

为了追踪人工智能政治广告支出的增长情况，CNN分析了今年播出的数千条广告，这些广告由追踪服务AdImpact收集。记者利用人工智能技术提取了每条广告屏幕上显示的所有文本，然后人工审核了那些包含相关披露信息的广告。分析结果显示，数十条广告向观众披露了其包含人工智能生成、增强或篡改的内容。

记者们还审查并发现了数十则未披露使用人工智能的广告，这些广告明显使用了人工智能生成的图像。其中一些广告被卫斯理媒体项目的研究人员标记出来。记者们还排除了一些无法确定使用了人工智能还是更传统的图形特效技术的广告。

分析发现，在广播电视政治广告中明确提及人工智能的支出，从 1 月份的不到 25,000 美元增长到仅 9 月上半月就超过 1000 万美元。

CNN 的分析很可能遗漏了一些使用人工智能但未披露的广告，因为这项技术的使用并不总是显而易见的。

上个月，共和党超级政治行动委员会“参议院领导基金”制作了一则攻击奥索夫的广告，该广告利用人工智能图像嘲讽他在竞选期间花费在发型和化妆上的钱。

广告中有人声称：“花了数千块钱做头发和化妆。真是个骗子。”广告中描绘了奥索夫接受美容护理，包括吹干头发、敷黄瓜眼膜和修指甲。

佐治亚州民主党人在广告发布后不久召开新闻发布会，呼吁撤下该广告，称“看到这些利用深度伪造视频制作政治广告令人深感不安”。SLF发言人则表示，真正“令人深感不安”的是奥索夫的过往记录。

德克萨斯州总检察长肯·帕克斯顿上个月推出的另一则广告显示，他的民主党参议员竞选对手塔拉里科在圣经上画画，并发表了有关宗教和性别的言论。

这则广告引用了塔拉里科多年来说过的一些话，但没有提供任何背景信息——而且这些话并不是由真正的塔拉里科说的，而是由人工智能生成的他的语音。

“肯·帕克斯顿和他在华盛顿的亿万富翁支持者们知道，他们不能依靠事实赢得这场竞选——所以他们唯一的办法就是用人工智能深度伪造视频，让詹姆斯说出一些话来试图欺骗德州人，”塔拉里科的发言人JT·恩尼斯在一份声明中说。

帕克斯顿竞选团队驳斥了这些批评。

帕克斯顿发言人麦迪逊·瑟西在一份声明中表示：“我们的团队一丝不苟地确保广告中的每一项内容都来源可靠、引用恰当。我们将继续利用一切可用手段，让德州民众了解詹姆斯·塔拉里科的激进记录。”

今年早些时候，在肯塔基州第四国会选区共和党众议员托马斯·马西的激烈竞选中，人工智能广告也成为了一个焦点。

反对马西的外部团体之一投放了广告，广告中使用了人工智能生成的图像，画面中这位国会议员与进步民主党众议员亚历山大·奥卡西奥-科尔特斯和伊尔汗·奥马尔手牵手。这两位进步民主党人虽然受到保守派的诋毁，但在反对美国对以色列的军事援助方面与马西立场一致。

广告中出现了疑似狗仔队拍摄的三人共进晚餐的照片，以及他们入住酒店的监控录像，旁白宣称马西“背叛了‘美国优先’运动，与‘四人帮’成员沆瀣一气”。广告开头几秒出现的一小段文字说明道：“本讽刺广告由人工智能生成。”

今年早些时候，马西在党内初选中落败后，他暗示这则广告可能是导致他失败的原因之一。“很多老年人投票反对我，因为他们认为我跟AOC和奥马尔约会过，”这位国会议员在X网站上发帖谈到这则广告时写道。“我跟AOC说过，我永远不会跟奥马尔那样！”他开玩笑说。

除了那些更具争议的例子之外，许多候选人也开始利用人工智能工具来制作争议较少的竞选内容，既简单又经济。共和党佐治亚州州长候选人里克·杰克逊今年在广告上花费了数百万美元，其中包括用人工智能制作的家庭照片动画和他艰苦童年生活的戏剧化演绎，这使得他的竞选活动成为CNN统计中人工智能广告支出最高的竞选活动。（虽然他的一些广告披露使用了人工智能技术，但另一些则没有——佐治亚州并不强制要求披露。）

研究人工智能在选举中应用的斯坦福大学法学教授纳撒尼尔·珀西利表示，深度伪造人工智能广告最终可能产生的最大影响，并非在重要的全州选举中，而是在选民可能不会花时间研究候选人或核实他们看到的广告的低级别选举中。

“从古至今，我们一直都有Photoshop等图像处理技术，以及政客们编造的谎言，”珀西利说。“但逼真的视频却能更容易地打动人心，也更具情感感染力。”

据全国州议会联合会统计，至少有31个州已通过法律，规范人工智能内容和深度伪造技术在政治广告中的使用，其限制措施从完全禁止到信息披露要求不等。这些法律大多是在过去三年内签署的，当时人工智能的应用日益普及。

大多数州都要求对政治广告中的人工智能内容或深度伪造技术进行某种形式的披露，而马里兰州、明尼苏达州和德克萨斯州则禁止了部分深度伪造广告。明尼苏达州和德克萨斯州的禁令仅适用于选举前的特定时期。

“今年是一个试验年，看看这些法律是否有效，是否得到执行，人工智能是否真的成为选举中的一个重要因素，”全国州议会联合会选举项目主任凯蒂·欧文斯·胡布勒说。

近几个月来，在中期选举中播出的一些人工智能广告似乎违反了这些法律。

在田纳西州，尽管该州在 7 月份通过了一项法律，要求在深度伪造政治广告中披露相关信息，但多个竞选活动仍然遭受了人工智能攻击广告的攻击。

该法案通过仅几周后，一个反对共和党州长候选人玛莎·布莱克本的“黑钱”组织播放了一则未经披露的AI广告，广告中布莱克本与制药公司游说者勾肩搭背，抹黑她为“制药女王”，并将她与该州的阿片类药物死亡事件联系起来。布莱克本的竞选团队向该组织发出停止侵权通知函，称该广告虚假且诽谤，并成功促使多家电视台停播了该广告。

在宾夕法尼亚州第六选区，两位共和党候选人也展开了激烈的交锋，双方都投放了人工智能驱动的广告。约翰尼·加勒特投放了一则广告，广告中一个佩戴着其竞争对手范·希拉里姓名标签的人从迈克尔·布隆伯格手中接过一大笔钱；而希拉里的竞选团队则以一则广告回击，广告中加勒特在办公室睡觉，还参加了骄傲游行。

双方竞选团队都向当地媒体表示，他们认为新的深度伪造法不适用于联邦公职候选人。

与此同时，在密歇根州，埃尔-赛义德的竞选团队强烈批评了这则人工智能制作的广告，该广告试图将这位候选人与穆斯林兄弟会和左翼 Twitch 主播 Piker 紧密联系起来。

负责投放这则广告的超级政治行动委员会“前线行动”（The Front Line Action）的发言人为该广告辩护。该发言人在一份声明中表示：“这则广告的内容基于埃尔-赛义德本人的公开声明，以及有关他本人和与其竞选活动相关或支持他的个人的书面报道和记录。”

争议点还在于：该广告适用哪条法规。

密歇根州的一项法律规定，政治广告中必须包含三秒钟的旁白，承认使用了人工智能技术，并附带四秒钟的书面披露。然而，另一项针对“具有实质性欺骗性的媒体”的选举法也规定，如果视频“虚假地描绘了所描绘的人物”，意图损害候选人的选举机会，且在选举前90天内发布，则必须在视频中全程披露相关信息。此外，广告发布者必须明知广告虚假地描绘了所描绘的人物。一位政治行动委员会（PAC）发言人表示，该组织认为第一项法律适用于其广告，并称该广告符合规定。

禁止传播欺骗性人工智能生成的广告或强制披露广告中使用人工智能的努力，在国会或联邦选举委员会中均未取得进展，而联邦选举委员会目前也因人数不足而无法运作。

法律专家表示，监管政治言论面临诸多障碍，尤其是当广告内容反映了目标候选人实际说过的话时。

左倾消费者权益倡导组织“公众公民”（Public Citizen）的州人工智能治理与技术主管伊拉娜·贝勒表示，通常情况下，可被解读为讽刺或戏仿的人工智能生成的广告不受州政府信息披露要求的约束。该组织一直倡导在联邦和州层面进行监管。

夏威夷州和加利福尼亚州的联邦法官裁定，这两个州的部分选举深度伪造法律违反了宪法第一修正案，属于违宪。

但即使没有联邦行动，越来越多的州级法规似乎也有助于限制贝勒所说的“噩梦般的场景”，即真正具有欺骗性、看起来很逼真的广告，在这种广告中，“公众看到某些东西，就会相信候选人以某种方式积极参与了道德上错误或丑闻的事情，从而影响公众舆论。”

撇开争议不谈，人工智能已经产生了广泛的影响，降低了政治广告制作的准入门槛。

政治广告商和观察人士表示，广告制作商一直在利用人工智能来简化流程，通过测试人工智能版本的广告来节省金钱和时间，如果获得批准，这些广告就会投入实际制作。

参议院多数党政治行动委员会（Senate Majority PAC）是民主党的一个主要团体，其通讯主管劳伦·弗伦奇（Lauren French）表示，虽然一些政治竞选活动一直以来都利用广告散布谎言，但他们现在“利用了以前无法使用的工具”。她批评攻击塔拉里科（Talarico）和奥索夫（Ossoff）的广告误导选民。

“我们仍然需要接受并利用这项新技术，”她谈到这项新技术时说。“但我们可以在使用它的同时保持道德操守，说真话。”

与此同时，其他一些竞选活动播放的广告中包含对手的真实视频，并特意声明这些视频并非由人工智能生成——试图帮助观众辨别真伪。

“我要给你们展示一些东西，没有人工智能，没有特效，只有迈克尔·沃特利自己的话，”民主党参议员候选人罗伊·库珀在今年早些时候播出的一则攻击性广告的开头这样说道。

斯坦福大学教授珀西利表示，选民最终可能会对人工智能攻击习以为常，甚至开始怀疑真实的视频。

他说：“选民将越来越难以分辨真假。这才是对民主更大的危害。”

CNN的杰伊·布朗对本报道亦有贡献。