An AI duck video sums up the problem: Van Orden mocks deepfakes after rival's legal ultimatum

By Zendoric · October 2, 2026.

Lawyers for Rebecca Cooke, the Democratic candidate in Wisconsin's 3rd District, sent a cease-and-desist letter on September 30 to Republican Representative Derrick Van Orden, according to documents cited by Wisconsin Public Radio. The reason: several videos made with "Grok Imagine", xAI's AI image and video tool, that Van Orden has posted on his X account showing a synthetic Cooke saying things she never said. In one, her avatar declares "let the peasants eat cake, they'll vote for me anyway"; in another, she says she has just voted "for two communists".

The letter, signed by attorney Ben Stafford, focuses on an August 4 video in which the AI-generated Cooke says she is running for Congress to "cash in on as much corruption as possible and give free health care to every illegal immigrant". The lawyers argue those statements are defamatory under Wisconsin law, which bars a public figure from acting with "actual malice" to damage another person's reputation. Van Orden responded publicly with another AI video: an animated duck shouting "Cease and desist!". Van Orden's campaign, for its part, accused Cooke of throwing a "tantrum".

The first thing worth stressing, because the article itself does, is that these videos are not undetectable: X automatically labels them as AI-generated. This is not a covert deception operation but a deliberate, openly declared use of synthetic satire as a campaign weapon, on the premise that the label supposedly absolves its user of liability. It is important to note that the defamation claim is, for now, just that: an accusation by the complaining party, not a fact proven in court.

Here is the real value of this episode, beyond the specific electoral context: it exposes the regulatory gap in real time. The "AI-generated content" label protects against the charge of concealment, but it does not settle whether the content is defamatory. These are two distinct legal questions, and almost no state framework has yet separated them clearly. Cooke's letter invokes common-law defamation precisely because, as she herself notes, "Wisconsin's statutory law against fake AI" leaves gaps. It is an implicit acknowledgment that rules written specifically for political deepfakes are not enough on their own.

Our reading is that disputes of this kind will multiply in every election cycle between now and 2028, not because the technology is going to improve —it is already effective and cheap— but because the legal and reputational cost of using it remains low. As long as an automatic platform label serves as a shield against accusations of defrauding voters, campaigns have an incentive to test the boundary between political parody (protected) and defamation (not protected), and to let a court decide, months after the election, where the line was. By then, the reputational damage will already have done its work in November.

Broadly, deepfake detection and synthetic-content labeling are advancing faster than the case law that should accompany them. That lag is not a technical problem with generative AI but an institutional vacuum: state legislatures need to define precisely what synthetic political conduct crosses from satire into defamation, before the precedent set by one isolated case ends up setting the rule for everyone.