\n Senate Republicans released an online ad this week in which a real-looking but fake version of a Democratic candidate, fabricated with artificial intelligence, appears to speak directly into the camera for more than a minute.\n <\/p>\n
\n The National Republican Senatorial Committee\u2019s deepfake of James Talarico, the Democratic nominee<\/a> in the US Senate race in Texas, is only the latest in a series of AI-generated creations from the national GOP campaign organization<\/a> in the past year. But it\u2019s the first featuring a phony version of a candidate talking in a lifelike manner for so long \u2013 an example of how far AI technology has come in a short time and an indicator of the direction attack ads may be heading.\n <\/p>\n \n \u201cThe face and voice are very good. There is a slight misalignment between audio and video, but otherwise this is hyper-realistic and I don\u2019t think that most people would immediately know it is fake,\u201d Hany Farid<\/a>, a University of California, Berkeley professor specializing in digital forensics, said in an email.\n <\/p>\n \n The use of AI deepfakes in campaign advertising raises a host of ethical questions. It has also prompted some bipartisan calls for federal legislation or regulation on the practice, though those ideas have also faced pushback on First Amendment grounds<\/a>.\n <\/p>\n \n The 85-second ad<\/a> depicts an AI-created \u201cTalarico\u201d appearing to proudly read excerpts from 2021 tweets in which the real Talarico made statements on transgender issues, race and religion and a 2013 tweet in which he recalled having attended a Planned Parenthood event as a teenager. In addition, the ad depicts the fake \u201cTalarico\u201d making new self-praising comments that there is no evidence the real Talarico actually made \u2013 praising the past tweets by \u201csaying\u201d things like \u201coh, this one is so touching\u201d and \u201coh, I love this one too.\u201d\n <\/p>\n \n The ad begins and ends with a narrator describing it as a \u201cdramatic reading,\u201d and an \u201cAI GENERATED\u201d disclosure appears on screen for almost the entire ad. But the disclosure text is small, mostly faint and confined to a bottom corner of the screen \u2013 and the fake \u201cTalarico,\u201d shown wearing a blazer and open-collared shirt, looks uncannily like the actual candidate.\n <\/p>\n \n A source familiar with the NRSC\u2019s thinking described AI as a \u201cconsistently effective\u201d\u202fway to highlight opposing candidates\u2019 statements and said, \u201cThese are Talarico\u2019s real words \u2026 all we have done is visualize them for voters using a modern tool, within all legal and ethical parameters.\u201d The source, however, said they did not have a comment to offer on the additional \u201cTalarico\u201d commentary the ad appears to have invented.\n <\/p>\n \n NRSC communications director Joanna Rodriguez asserted in an email that Democrats are \u201cpanicking after hearing James Talarico\u2019s own words.\u201d Talarico campaign spokesperson JT Ennis asserted in a text message that it is the candidates in the ongoing Republican primary who are \u201cscared of James Talarico,\u201d adding, \u201cWhile they spend their time making deepfake AI videos to mislead Texans, we are uniting the people of Texas to win in November.\u201d\n <\/p>\n \n Texas has one of the country\u2019s strictest state laws on political deepfakes, but it only applies in the month prior to an election. A Texas bill<\/a> passed in 2019<\/a> made it a criminal misdemeanor, punishable by up to a year in jail, to create a deepfake video and cause it to be published or distributed within 30 days of an election if it was \u201ccreated with the intent to deceive\u201d and intended to hurt a candidate or influence the results.\n <\/p>\n \n Election Day in the 2026 midterms is in early November, while it\u2019s late May for the Republican primary runoff. And while roughly half<\/a> of states have passed a law relating to campaign deepfakes, many of the others simply require disclosure when an ad was made with AI. Democratic Sen. Andy Kim of New Jersey responded to the anti-Talarico ad with a call for national action, posting<\/a> on X: \u201cThese deepfakes are dangerous and wrong. We need protections not just for politics, but for all Americans that could be targeted.\u201d\n <\/p>\n \n The words \u201cAI GENERATED\u201d appear in small text in the bottom right corner of the anti-Talarico ad, above the NRSC logo, for about three seconds soon after it begins. Then, as the fake \u201cTalarico\u201d speaks, the words \u201cAI GENERATED\u201d appear in fainter and even smaller text in the same corner, staying on the screen for more than a minute. The slightly larger and darker disclosure text reappears for the final five seconds of the ad.\n <\/p>\n \n Some previous AI use by political campaigns did not include any disclosure \u2013 for example, in 2023, when the Republican presidential campaign of Florida Gov. Ron DeSantis posted<\/a> fake images of President Donald Trump hugging Dr. Anthony Fauci mixed in<\/a> with real images, or a 2024 robocall scandal in which a consultant working for the Democratic presidential campaign of Rep. Dean Phillips hired someone<\/a> to create an AI version of President Joe Biden\u2019s voice urging New Hampshire voters not to vote in the primary.\n <\/p>\n \n Sarah Kreps<\/a>, professor and director of the Tech Policy Institute at Cornell University, said the disclosure in the ad against Talarico \u201creflects the direction the technology and the norms around it seem to be moving.\u201d\n <\/p>\n \n \u201cCampaigns seem like they\u2019re starting to treat synthetic media less as something covert \u2013 maybe because that backfired as being dishonest and deceptive, not qualities you want to see in elected officials \u2013 and more as something that can be used openly as long as viewers are told what they\u2019re seeing,\u201d Kreps said in an email.\n <\/p>\n \n It\u2019s debatable, though, whether the small-text disclosure was truly open.\n <\/p>\n \n \u201cI don\u2019t think that faint, small font in the bottom righthand corner comes close to appropriate disclosure because the average person doom scrolling on X\/YouTube is simply not going to notice \u2014 in fact, I didn\u2019t notice when I first looked at the video,\u201d said Farid. \u201cI also think that if the (Talarico) tweets are real, seeing the candidate read them lands differently and can reasonably be categorized as deceptive, and I don\u2019t think that the campaigns or candidates should be opening this Pandora\u2019s box.\u201d\n <\/p>\n \n AI fakery has proliferated during the 2026 midterm cycle as rapid improvements in AI technology have made fake videos more convincing and easier to create than ever<\/a>.\n <\/p>\n \n Texas is a prime example. As The Texas Tribune noted<\/a> last month, multiple ads and social media posts have employed AI videos and images during the contentious Republican Senate primary involving Sen. John Cornyn and Texas Attorney General Ken Paxton.\n <\/p>\n \n An attack ad<\/a> from Paxton\u2019s campaign made extensive use of a fake \u201cCornyn\u201d happily dancing with Democratic Rep. Jasmine Crockett; small text at the end of the ad disclosed that \u201ccertain video\u201d in it was AI \u201csatire that does not represent real events.\u201d An ad<\/a> from Cornyn\u2019s campaign, meanwhile, showed phony clips of unsuccessful Republican candidate Rep. Wesley Hunt holding a Pomeranian in fake scenes to depict Hunt as a mere \u201cshow dog\u201d; that ad did not include any AI disclosure.\n <\/p>\n \n Various Democrats have used AI, too. California Gov. Gavin Newsom has posted<\/a> a fictional video showing Trump and top administration officials crying in handcuffs as well as other AI content, though much of it clearly fake and satirical<\/a>. Crockett\u2019s unsuccessful Senate primary campaign would not answer directly when Texas media<\/a> asked<\/a> if an ad\u2019s striking image of Crockett standing with a massive crowd was AI-generated.\n <\/p>\n \n The NRSC, like others, sees little downside to deepfakes in ads. Even when some viewers express outrage or news outlets cover the story of a fake, the ad \u2013 and the messages the ad is trying to highlight \u2013 get more attention. Kreps said synthetic media is \u201clikely to become a routine campaign tool\u201d in both parties.\n <\/p>\n \n \u201cWhat we\u2019re likely seeing is a kind of competitive boundary-pushing: once one campaign demonstrates a tactic, others adopt it rather than risk a perceived disadvantage,\u201d she said.\n <\/p>\n