[Image: X / Deedy Das]
OpenAI’s shiny new 4o image generator model is freakishly good at something that’s tripped up its clunky ancestors for years: putting actual legible text inside images.
And that, dear reader, makes it an absolute gift to anyone with a taste for mischief and fraud.
Turns out, it’s stupidly easy to whip up fake documents that look legit enough to fool your gran, your boss, or probably even your bank.
Case in point: Deedy Das, a principal at Menlo Ventures, fired off a cheeky tweet featuring a very official-looking receipt for a decadent meal at a real San Francisco steakhouse – all whipped up with 4o, as TechCrunch noticed.
You can use 4o to generate fake receipts.
There are too many real world verification flows that rely on “real images” as proof. That era is over. pic.twitter.com/9FORS1PWsb
— Deedy (@deedydas) March 29, 2025
And he’s not wrong. At first glance, the image is slick with a detailed breakdown of the meal, subtotal and tip all neatly in place. Basically, something your company’s finance team would greenlight without blinking.
Another user took it even further, slapping on a greasy filter and food stains to complete the illusion. Gourmet expense fraud, anyone?
But wait, it gets darker. Das also tested 4o’s limits by asking it to generate fake prescriptions for controlled substances like Zoloft – and, surprise, it obliged.
check this skill, recreating ticketshttps://t.co/tXGKEu45I5
— Rahul Chakraborty (@hckmstrrahul) March 29, 2025
Enter the Fraud Squad![™]()
This isn’t just a party trick. It’s a sign that AI-generated imagery has levelled up hard. Not too long ago, these models were vomiting out mangled letters and gibberish that looked like a ransom note written by a drunk alien.
Now? They’re borderline criminal masterminds. Fake tax forms. Bogus bank cheques. DIY IDs and birth certs. If it’s a document, 4o probably won’t flinch.
Whether we’re remotely prepared to spot these AI-spawned fakes is another story entirely. So far, attempts to slap on metadata or watermarks as “proof” that an image is AI-generated are laughably easy to sidestep.
Oh, and it’s not like fraud was born yesterday. Back in 2015 – pre-AI apocalypse – a survey found that 85% of people happily fibbed their way to fatter reimbursements. Most of it flew under the radar, thanks to sloppy internal controls and janky accounting systems.
So yeah. The line between what’s real and what’s cooked up in silicon? It was blurry and now it’s basically gone.
Welcome to the age of believable lies.
[Image: Futurism]