How OpenAI got hacked with an image

Two guys hacked OpenAI with a malicious HEIF image. They turned a one year old libheif heap overflow into a remote code execution on OpenAI's Discourse forum, took over an employee's ChatGPT account, and left a message in the internal monorepo. This story shows how AI changes the economics of exploit development and why security through complexity doesn't work anymore.

LEARN ON HEXTREE (ad)
Learn hacking on Hextree: https://www.hextree.io/
Learn more about AI on Hextree: https://app.hextree.io/map/artificial-intelligence
Join the Hextree Discord: https://discord.gg/xgQpCQCpvy

IMPORTANT FOR SELF-HOSTED DISCOURSE
Rebuild your Discourse installation to pull the current image-processing dependencies. Updating only through the web interface may not replace an older Libheif package.

RESOURCES
Hacktron research: https://www.hacktron.ai/blog
Harsh on X: https://x.com/rootxharsh
Mohan on X: https://x.com/S1r1u5_

CHAPTERS
00:00 - Introduction
01:37 - Chapter 1: Vulnerability Research
05:39 - Chapter 2: The Vulnerability in Libheif
08:49 - Chapter 3: Hacking into OpenAI
11:32 - Chapter 4: Disclosure
13:17 - Chapter 5: Hack the Planet
17:19 - Outro

SUPPORT
Per video: https://www.patreon.com/join/liveoverflow
Per month: https://www.youtube.com/channel/UClcE-kVhqyiHCcjYwcpfj9w/join
Buy my handwriting font (ad): https://shop.liveoverflow.com/

WATCH, FOLLOW & READ
Second channel: https://www.youtube.com/LiveUnderflow
Twitch: https://twitch.tv/LiveOverflow/
Twitter: https://twitter.com/LiveOverflow/
Instagram: https://instagram.com/LiveOverflow/
TikTok: https://www.tiktok.com/@liveoverflow_
LiveOverflow blog: https://liveoverflow.com/
Hextree blog (ad): https://www.hextree.io/blog

#OpenAI #ChatGPT #LiveOverflow

(ad) LiveOverflow YouTube channel is supported by advertisement and product placement.