Executive Summary

AI deepfakes in the 2026 midterm elections moved from hypothetical to routine. In Wisconsin's 3rd District, Republican Rep. Derrick Van Orden posted AI-generated videos of his Democratic challenger, Rebecca Cooke, saying things she never said, and Cooke went on to win on November 3, 2026, in a race CBS News projected and Van Orden conceded. In Maine, a fake TV newscast targeted Democratic Senate candidate Troy Jackson; NewsGuard says it appears to be the work of the Russian operation Storm-1516, and Jackson also won. Whether any deepfake changed a single vote is not established.

A Duck, a Letter and a Congressional Race

On September 30, 2026, lawyers for Rebecca Cooke sent a cease-and-desist letter to the sitting congressman she was trying to unseat. Their complaint was specific: Rep. Derrick Van Orden had posted AI-generated videos of Cooke on X, showing her saying things she never said. "Ms. Cooke never made the statements attributed to her," wrote her attorney, Ben Stafford, according to Wisconsin Public Radio.

Van Orden answered on X with another AI video. WPR reported it showed a red, AI-generated duck exclaiming "Cease and desist!" Axios reported that the duck wore Democratic Socialists of America apparel. The Milwaukee Journal Sentinel reported that he also urged Cooke to file a mock "hurt feelings report."

Five weeks later, Cooke won the seat. Her race is the clearest documented case of how AI deepfakes entered the 2026 midterm elections: openly, from a candidate's own account, and with an AI watermark attached. It also shows how little the law could do about it.

What Happened in the Van Orden–Cooke Deepfake Fight?

August: the first videos

The videos began in the summer. One post from August paired an AI-generated Cooke with an AI image of former President Joe Biden. In it, the fake Cooke says she is running for Congress to "take advantage of as much corruption as possible and give every illegal alien free healthcare," as Axios and WPR quoted it. WPR listed other invented lines, including "Let the peasants eat cake. They will vote for me anyway."

The Journal Sentinel described other clips: Cooke pretending to type on an oversized calculator, calling herself "the anointed one," and being mocked by a cow during dairy month. WPR reported that Van Orden made the videos with Grok Imagine, the image and video generator from xAI. The Journal Sentinel reported that the videos typically carried a Grok watermark when they appeared on X.

Cooke's public response was direct. "Derrick Van Orden can't run on his record, and is instead spending his time and energy producing AI videos putting words in my mouth," she said, according to WPR. Van Orden's campaign answered: "Rebecca Cooke has spent six years running for office, lying to voters about who she is and her record."

Late August: a legal gap opens

Wisconsin has a law on AI in campaign ads. But on August 27, Spectrum News reported that Van Orden's posts appeared to violate its "spirit" rather than its text. Bruce Boyden, an associate professor at Marquette University Law School, told Spectrum: "I think there's probably a good argument that the law only covers when you pay for particular advertisements using campaign contributions."

September 30: the letter, and the duck

Cooke's letter relied on defamation rather than the election statute. Stafford argued that "Wisconsin's common law of defamation fills the gaps yet remaining in Wisconsin's statutory protections against fake AI," WPR reported. Van Orden's campaign called the letter a "stunt," according to CBS News. We found no report of a lawsuit as of November 5, 2026.

October: an ethics complaint, and a dispute over a photo

On October 22, an Eau Claire resident filed a complaint with the Wisconsin Ethics Commission, WEAU reported. It alleged that an October 19 graphic posted by the @TeamVanOrden account used an AI-generated image of Cooke without the disclaimer required by Wis. Stat. § 11.1303(2m).

Van Orden disputes this. On October 23, he said the image is "a photograph of Rebecca Cooke taken during a May interview with News 8." He said the original footage proves it is real. We found no public ruling on the complaint as of November 5, 2026.

November 3: the result

CBS News projected early on November 4 that Cooke had ousted Van Orden, a two-term incumbent. It was her third run for the seat; she narrowly lost the 2024 general election. Van Orden conceded on election night. "Now she's going to represent me in Congress," he told supporters, urging them to support her, according to Spectrum News. "We did this," Cooke told her own supporters.

The result is a media projection plus a concession. It is not yet official. Wisconsin's results become official only when the state canvass is certified, due by December 1, 2026.

Who Made the Fake Troy Jackson Newscast in Maine?

In the last week of the campaign, a different kind of deepfake appeared in Maine. It looked like a news report from WABI-TV, a Bangor station. In the clip, a newscaster's voice claims that Troy Jackson, the Democratic Senate candidate, threatened to kill "any insubordinate staffer" with an axe, Mashable reported on October 30.

It was not real. WABI vice president Kim Petry said the video "was not produced, aired, or authorized by WABI," and confirmed that it used synthetic audio. NewsGuard found that the fake combined dubbed audio with genuine WABI footage from July 2026, Mashable reported. That footage covered Jackson's documented outbursts as Maine Senate president. Jackson has acknowledged throwing a water bottle in 2021 but denied aiming it at anyone, and he never threatened anyone with an axe.

The attribution to Russia is NewsGuard's assessment. NewsGuard, a company that rates the reliability of news sources, described the video as apparently produced by Storm-1516, a Russian influence operation that, in its words, "routinely fabricates articles and videos that impersonate legitimate media outlets." We could not load NewsGuard's own Reality Check post (the site returned an access error), so its findings here come from Mashable and PolitiFact.

PolitiFact reported on October 30 that other researchers also linked the campaign to Storm-1516. They included Joseph Bodnar of the Institute for Strategic Dialogue, Clemson University professor Darren Linvill and the analytics firm Graphika. The same researchers tied similar fake videos to the network. Those videos falsely accused two other Democratic Senate candidates, James Talarico in Texas and Sherrod Brown in Ohio, of sexual misconduct. None of the sources we reviewed cited a US government attribution.

Jackson won anyway. The Associated Press called the race at 1:02 a.m. on November 4, after Susan Collins had already conceded, Maine Public reported. Jackson was leading by seven points with 75% of precincts reporting.

Are AI Deepfakes of Political Opponents Legal?

There is no single answer, because there is no federal law that settles it. Rules vary by state, and courts are narrowing them.

Wisconsin's law

Wisconsin's statute, Wis. Stat. § 11.1303(2m), came from 2023 Wisconsin Act 123. It defines "synthetic media" as audio or video "substantially produced in whole or in part by means of generative artificial intelligence." Video communications covered by the law must display "This video content generated by AI" while the synthetic portion is on screen. An intentional violation carries a forfeiture of up to $1,000 per violation.

The catch is scope. The requirement applies to communications paid for with contributions or disbursements, which is why legal experts doubted it reached a candidate's unpaid social media posts. The Journal Sentinel reported that the law does not technically apply to social media posts as opposed to formal ads. A "Made with Grok Imagine" watermark is not the disclaimer the statute prescribes, but the statute may simply not apply.

Other states and the courts

About 31 states regulate political deepfakes, according to Public Citizen's tracking as reported by Pluribus News in the Arizona Capitol Times. Most require disclosure within a set period before an election. Minnesota and Texas prohibit distribution in certain pre-election windows, and Maryland bans such deepfakes year-round.

Courts have pushed back. Judges have struck down election deepfake laws in California and Hawaii on constitutional grounds, Pluribus News reported, though a California labeling law remains in effect. On September 18, 2026, US District Judge Susan Watters ruled that Montana's law violates the First Amendment, the Daily Montanan reported. Her ruling protects only the plaintiffs, a PAC treasurer and his committee.

Federal agencies

The Federal Election Commission declined in September 2024 to write AI-specific rules. Instead it adopted an interpretive rule stating that the ban on fraudulently misrepresenting campaign authority "is technology neutral," including for AI-assisted media. The Federal Communications Commission fined the consultant behind the 2024 fake Biden robocall in New Hampshire $6 million, although he was acquitted of criminal charges. Congress has set no federal standard for AI in political ads, Axios reported. Federal executive agencies now use "Super Intelligence" or "SI" in place of "AI" under a September 2026 executive order, Crowell & Moring reported, but no new election rule has followed.

Does an "AI-Generated" Label Make a Deepfake Acceptable?

The Wisconsin videos typically carried a tool watermark. A viewer who noticed it could tell the clip was synthetic. The words were still invented, and they were still attributed to a real candidate. That is the ethical tension of this cycle: a label can be present while the message remains deceptive.

The platform policies described to CNBC address paid ads rather than ordinary posts. YouTube told CNBC that "we require disclosure when an election ad contains synthetic or digitally altered people or events." Meta uses "paid for by" labels, an ad library and an "about this ad" tool for ads detected as AI-made. X says advertisers must comply with applicable laws, which places responsibility on the advertiser. We found no public statement from X or xAI about the Van Orden videos.

The scale is large. CNBC reported that the Wesleyan Media Project counted more than $187 million spent on nearly 300 AI-generated or AI-enhanced ads this cycle, and that most lacked AI disclaimers. These are Wesleyan's figures as reported by CNBC; other reports at earlier dates gave lower totals.

AI companies also acted. OpenAI opened an "Election Operations Center" to run until all votes were counted, led by Josh Lawson, its head of civic and strategic risk, Cox Media Group reported. Lawson said OpenAI converts photorealistic election image requests into cartoons "so it can't be used for deceptive purposes." There is no public data yet on how well these safeguards worked. They also apply only to OpenAI's tools; the Wisconsin videos came from a different company's product.

Did AI Deepfakes Change the 2026 Midterm Elections?

We found no evidence that they did. The two best-documented targets, Cooke and Jackson, both won. TIME's November 4 analysis concluded that deepfakes "do not appear to have had an outsize impact in many races." That is analysis, not a measured finding. We found no study that has yet tested whether voters who saw these videos changed their minds.

AI was still on the ballot in another sense. TIME reported that 56% of voters in an NBC exit poll called AI and data centers a "crucial or major factor" in their vote for Congress. CBS reported that both Cooke and Van Orden accused each other of supporting pro-data center policies.

Claims that the Wisconsin videos "backfired" on Van Orden are speculation. Cooke had a fundraising advantage, nearly $12 million to Van Orden's $8.8 million according to FEC data cited by CBS, and many issues were in play.

What Happens Next?

  • November 10, 2026: Wisconsin county boards of canvassers must convene to begin official canvasses, according to the Brennan Center's certification guide.
  • November 17, 2026: Wisconsin counties must deliver certified results to the Wisconsin Elections Commission.
  • By December 1, 2026: The commission chair must certify Wisconsin's results, making Cooke's win official.
  • Pending: The Ethics Commission complaint over the October 19 graphic has no announced outcome.
  • Ongoing: Court challenges to state deepfake laws will shape what states can enforce before the 2028 elections.
  • January 3, 2027: The new Congress convenes, with Cooke and Jackson expected to be sworn in.

AI Coalition Network Perspective

The 2026 midterms showed two separate deepfake problems that need separate answers. The first is domestic and open: a named candidate posting labelled synthetic speech of a named opponent. Disclosure laws tied to campaign spending did not clearly reach it, and defamation law is slow and expensive. Legislators who want to cover organic posts will have to do so within the limits that courts in Hawaii, California and Montana have now set.

The second problem is covert and foreign: an anonymous operation that impersonates a trusted local news brand. No disclaimer law reaches that. Fast debunking by the impersonated outlet, researchers and fact-checkers did the work in Maine.

A watermark tells you how a video was made. It does not tell you whether the candidate said it. Platforms, campaigns and voters should treat "AI-generated" as the start of the question, not the answer.

Frequently Asked Questions

What did Derrick Van Orden post about Rebecca Cooke?

Starting in August 2026, Rep. Derrick Van Orden posted AI-generated videos on X showing his Democratic challenger, Rebecca Cooke, saying things she never said. WPR reported he used xAI's Grok Imagine. Cooke's lawyers sent a cease-and-desist letter on September 30 citing defamation. Van Orden's campaign called it a "stunt," and Cooke won the seat on November 3.

Are political deepfakes legal in Wisconsin?

Wisconsin Stat. § 11.1303(2m) requires paid political communications with AI-generated audio or video to carry a disclaimer such as "This video content generated by AI," with forfeitures up to $1,000 per violation. Legal experts told Spectrum News in August 2026 that the law likely covers only paid ads, not a candidate's unpaid social media posts.

Who made the fake Troy Jackson newscast?

The fake WABI-TV clip, reported on October 30, 2026, falsely claimed Maine Democrat Troy Jackson threatened staff with an axe. NewsGuard said it appears to be the work of Storm-1516, a Russian influence operation, and other researchers agreed, PolitiFact reported. That is a researcher assessment, not a government finding. WABI said it did not produce the video.

How many states regulate election deepfakes?

About 31 US states regulate political deepfakes as of mid-2026, according to Public Citizen's tracking. Most require disclosure before an election; Minnesota and Texas restrict distribution in pre-election windows. Courts have struck down election deepfake laws in Hawaii and California, and in September 2026 a federal judge blocked Montana's law as applied to one PAC.

Did deepfakes affect the 2026 midterm results?

No published evidence shows that deepfakes changed any 2026 result. The two best-documented targets, Rebecca Cooke in Wisconsin and Troy Jackson in Maine, both won on November 3, 2026. TIME's post-election analysis said deepfakes did not appear to have an outsize impact in many races, but no study has yet measured their effect on voters.

Sources