The first time I saw the screenshots, the room went quiet. Someone whispered that Kimberly Guilfoyle had texted a donor asking to cover a $100,000 American Express bill ($100,000; €92,000). Then a claim came through: the texts were AI-generated.
I want you to hold the scene: reporters, anonymous sources, a president saying he “didn’t like what we were reading.” I’ll walk you through why that single sentence—AI did it—was supposed to do so much heavy lifting, and why it apparently didn’t.
You can read a damning text in ten seconds. The timeline that followed made it worse.
The Wall Street Journal published screenshots of texts that, if authentic, looked like a direct ask: Kimberly Guilfoyle requesting help with an American Express balance ahead of a confirmation hearing. You remember the specifics—Signal, the donor Eric Deters, and a paper trail that spread across Newsweek, the New York Times, and cable punditry.
At that point, a lot of people do simple math: scandal + optics = headache. Guilfoyle’s alleged message—asking someone to pay off $100,000 (€92,000) on her behalf—was the kind of headline that forces quick decisions inside a campaign. President Trump, per PBS reporting, told reporters he wanted the story to stop; according to The Hill, GOP strategists said midterms made any distraction intolerable.
A White House phone call happened. The reaction tells you more than the claim did.
Someone in the West Wing decided the story needed containment. Guilfoyle, according to the WSJ, reached out to say the screenshots were AI-generated.
I’ve watched many excuses fly across tables in Washington. This one landed like a dropped stone in a pond. You can see why: in an era where images and texts can be manipulated, invoking AI offers plausible deniability and a neat cleanup. But anonymous White House sources told the Journal they didn’t buy it. That disbelief is as instructive as a confession.
Did Kimberly Guilfoyle claim the texts were AI-generated?
Yes, the WSJ reports she told White House officials the messages were produced by artificial intelligence. But read the context: the donor, Eric Deters, had circulated the screenshots himself, per Newsweek. If someone is distributing screenshots, saying “AI made them” is a narrow defensive move—one that invites a simple, testable counter: show the original files, request device logs, or ask the distributor how they obtained the images.
The supposed defense was short on inventiveness, long on implications.
Imagine you’re on the receiving end: a screenshot, an accuser, a named intermediary. Saying “AI” without a follow-up forensic thread is not a strategy; it’s a hope.
That hope works only when the audience prefers to suspend disbelief. If the person you’re telling is willing to play along, the label AI functions like a universal shrug. Here, it didn’t. The GOP, facing an all-hands election moment, opted for damage control rather than a messy public fight over authenticity.
Can AI create fake text message screenshots?
Yes—tools and techniques from image editors to neural-image generators can fabricate believable screenshots. Platforms like Signal and the American Express app are easy to mimic visually. But forgery leaves traces: metadata, message timestamps, the distribution chain. When reporters and officials can trace a screenshot back to a human distributor like Deters, the simple “AI did it” narrative starts to fray.
Reporters followed the chain. The chain led to questions, not absolution.
Journalists at the WSJ, Newsweek, and the Times pursued who shared what, when, and why. That’s how anonymous sourcing and attribution function as pressure tests—the story either holds under scrutiny or it collapses.
Guilfoyle resigned after the coverage intensified. Resignation didn’t equal admission or proof; it did, however, signal a political calculus: the optics had become too heavy for a campaign in midterm mode. Still, the refusal to accept a quick AI explanation suggests that even now, the invocations that once drowned out facts can be countered by the old-fashioned work of reporting.
The interesting part, to me, is not only that the claim was made but that it was rejected. If we’re living through an epistemological crisis, this episode reads like a small test case: how far will people let the phrase “AI-generated” absolve responsibility? On which side of that line are you willing to stand?