Hackers Broke Into OpenAI's ChatGPT Using a Photo — And Your Apps May Share the Flaw

A security team called Hacktron discovered they could sneak malicious code into OpenAI's public help forum (community.openai.com) simply by uploading a specially crafted image file. Think of it like a Trojan horse disguised as a photo — the forum's image reader didn't check it carefully enough. From there, they could hijack ChatGPT and Codex accounts, and because those accounts connect to GitHub, Slack, and email, the potential damage was enormous.

To prove the hack was real without actually stealing secrets, the researchers used an employee's coding assistant to open a harmless test file inside OpenAI's private code library. The whole thing — from spotting the flaw to proving access — took under 72 hours. OpenAI patched its side in roughly 14 hours, and the forum software maker, Discourse, had a fix ready within days. OpenAI paid a $6,500 bounty, noting the award covered only its own side of the issue.

The bigger warning is that this isn't just an OpenAI problem. The underlying bug lives in a popular image-processing library called libheif, which is bundled into software used by Slack, Meta, GitHub, and many web frameworks. If any app you use lets people upload .heic or .heif photos (the format iPhones default to), it may share the same weakness. The researchers spent months tracing how far this one library spreads.

So what should you do? You don't need to panic — most major companies patch quickly once alerted. But it's a reminder that convenience features like photo uploads are also attack surfaces. If you run a website that accepts user images, update your software now. And if you use ChatGPT or similar tools, consider what sensitive accounts you've linked to them, because one weak login can unlock a lot.