A social media user has sparked an online wave of alarm. The user, known as Shirish on X, shared a hyper-realistic AI-generated cheque.
It appeared to be an official UCO Bank cheque for ₹69,000. The cheque was made payable to Shirish's own name. He claimed it was created using OpenAI's newly-launched ChatGPT Images 2.0 model. His post carried a chilling caption: "We are so cooked."
The image went viral almost immediately. It showed every element of a real bank cheque. The account number, bank branch details, signature line and MICR band at the bottom all looked authentic. The level of detail alarmed thousands of users across the platform.
Social Media Reacts
TL;DR: Another said ChatGPT was not supposed to allow these outputs at all.
“This is a serious breach of law and policy. How can AI create a fake cheque?” wondered the user.
Another said ChatGPT was not supposed to allow these outputs at all. Some users remained sceptical about the actual fraud risk.
Banks require account holders to verify cheque details for cheques above a certain amount. Without real barcodes and account numbers, the cheque cannot be cashed.
According to one user, things will get truly dangerous when AI can generate working barcodes that banks accept. Another user reported that AI had already generated a working barcode from an existing one.
"So did you get the ₹69,000?" asked one user. Another joked that the real problem was only asking for ₹69,000.
“We are finished,” exclaimed another user.
Note: This report is based on user-generated content from social media. LiveMint has not independently verified the claims and does not endorse them.
ChatGPT Images 2.0: Are we ‘cooked’?
TL;DR: Until recently, AI image tools struggled badly with text inside images.
Until recently, AI image tools struggled badly with text inside images. Ask an AI to write words on a signboard, and you would get garbled letters. That is no longer the case.
ChatGPT Images 2.0 can now accurately render small, readable text within images. Think of a cinema poster with actor names clearly printed. Or, imagine a restaurant menu with legible prices and dish names. Or, it’s a fake boarding pass with a real-looking flight number and gate.
The model can generate working QR codes. Scan one, and it actually takes you somewhere. Earlier models produced decorative patterns that scanned as nothing.
It supports images up to 2K resolution. A fake identity card now looks sharp even when zoomed in.
It handles dense layouts accurately. Think prescription pads, government forms, official letterheads and bank documents, all look entirely genuine.
It reads and writes correctly in non-English languages. A fake Aadhaar-style document in Hindi or a Japanese official letter is now within reach.
It can search the web in real time while generating. A poster about a real upcoming event can include accurate dates and venue details.
OpenAI says the model is available to all ChatGPT users today. Thinking-based image features are available to Plus, Pro and Business subscribers.
So, are we ‘cooked’? That depends on who’s using it. Making an A-bomb and dropping it on Hiroshima are two different things.
About the Author
Stay updated with the latest Trending, India , World and US news.
Curated by James Chen
