Images purporting to show President Donald Trump kissing White House aide Natalie Harp on a golf course ricocheted about social media this week, amassing millions of views across online platforms.
They were fakes — circulated with no clear origin and bearing telltale visual signs such as strangely distorted human hands — that were whipped up using generative artificial intelligence tools, according to a Washington Post analysis using AI detection tools from ChatGPT-maker OpenAI and Google. (The Washington Post has a content partnership with OpenAI.)
Two versions of the smooching scene — one clear and one blurry — first surfaced online Monday, according to reverse-image searches conducted by The Post on Google and the website TinEye.
“This is clearly a fake photo,” the White House said in an email to The Post, before attributing it to a political hit job. “Of course this is coming from Democrats who have nothing of substance so instead they peddle fake images.”
While some people were certainly fooled — “Today is a bad day to have eyes,” lamented one Threads user — debunkers online were quick to try to correct the record. “Her hand has six fingers,” a commenter pointed out in response to a popular post on X. They included a zoomed-in close-up of the female hand showing an extra digit as well. (AI image models are notoriously bad at constructing hands.)
Despite the image’s obvious errors, its quick spread signals an increasingly common undermining of people’s trust, said Hany Farid, an expert on AI-generated images at Dartmouth College.
This is “death by a thousand cuts,” he said. “When you sow distrust of the visual record, nothing has to be real. Nothing’s fake, nothing’s real.”
One of the first viral shares came from an X user identified as Mykhailo Rohoza, who posted the clearer image on Monday evening.
“Thirty years ago, a photo like this could have destroyed a politician’s career overnight. Today? Barely worth interrupting the news cycle,” he wrote. “But don’t worry — I’m sure the guardians of ‘Christian family values’ are preparing their furious statements as we speak.”
In an email to The Post on Friday, Rohoza said he was surprised by his post’s virality, which he’d sent as a joke. “I certainly did not expect it to receive more than 10 million views,” he said.
Rohoza insists he did not create the image, and could not recall who posted it to the original thread, but says he slightly altered it using AI. “I found a blurry, low-quality version of it circulating on X. I used ChatGPT to brighten and enhance the image, but I did not add, remove, or alter the six fingers visible in it.” he wrote. “They were already present in the image I found.”
A popular left-wing influencer named Keith Edwards on Monday used the clearer image in a YouTube thumbnail for one of his videos — one of the earliest known uses. He didn’t use the image within the actual video, or discuss its contents, but it remains in use as the thumbnail with overlaid letters “WTF.”
“My thumbnail team found the image online,” Edwards told The Post in an email Friday. “We did not create or alter the underlying image, but we also did not independently verify its original source or that the woman pictured is Natalie Harp.” Edwards did not respond when asked where his team found the image.
The effort required to debunk the images and their variants online underscores the difficulty of moderating easy-to-make AI photos and videos on social media.
Most likely, the high-resolution image came first, Farid said, with the blurry version perhaps created by someone trying to get rid of the AI watermark. AI detectors from two different companies flagged the blurry photo, he said, indicating that someone may have tried to re-enhance that version.
Perhaps the most thorough deconstruction of the hoax came Thursday from BBC journalist and fact-checker Shayan Sardarizadeh, who wrote a nine-part thread on X, discussing his process for determining that both images were resolutely fakes.
After some digging, he found evidence that both images were created with AI. The clear version with the six-fingered Harp lit up an OpenAI detector, indicating that it was created with the company’s tools. The blurry version was trickier and, at first glance, did not generate flags from Google or OpenAI’s detectors.
Eventually he found a version that Google’s own AI detectors flagged. (Both OpenAI and Google watermark images generated with their tools like ChatGPT and Gemini, and allow users to check the origin of photos for themselves.)
Still, Sardarizadeh was not able to determine who first posted the image, and whether the blurry or the clear version came first.
“Verifying content in the age of AI is now becoming a complex, time-consuming, and [lengthy] process,” he wrote on X.
The post Images of Trump kissing Natalie Harp went viral. They’re AI-generated fakes. appeared first on Washington Post.
