Wes Roth · 2026-09-19 · notable
Wes Roth — 'OpenAI JUST got HACKED...'
Wes Roth posted 'OpenAI JUST got HACKED...' on 19 September 2026. The subject named in the title is Hacktron's bug-bounty write-up, which describes a chain that reached OpenAI's internal code repositories.
Wes Roth walks through the Hacktron bug-bounty chain that reached OpenAI's internal code, in a video posted on 19 September 2026.
What is it?
'OpenAI JUST got HACKED...' went up on the Wes Roth channel on 19 September 2026. The write-up behind it is Hacktron's 'Hacking OpenAI' post, which lays out a nine-step chain that ended with access to OpenAI's internal repositories. The bugs were reported through OpenAI's bug-bounty programme, OpenAI paid $6,500, and the issues are fixed.
How does it work?
The chain starts with a photo upload. Discourse, the forum software OpenAI's community site runs on, did not check HEIC/HEIF files itself, so it handed them to ImageMagick, which decoded them with libheif 1.19.7 — a version that still carried a heap buffer overflow in Debian 12. That gave code execution on the forum. A second flaw, in how OpenAI validates sign-ins, then turned forum access into access to employee ChatGPT and Codex accounts, and one of those accounts had Codex connected to OpenAI's GitHub organisation.
Why does it matter?
Exploit writing is the part of this story worth watching. Hacktron says Opus 4.8 struggled across several sessions to produce a working exploit, and that Claude Opus 5 succeeded within hours of its release, landing an ARM64 exploit in 3 hours. The full path from first bug to internal repository took under 72 hours. Roth's channel is where a lot of people will meet that detail, which matters to anyone running a forum, an image pipeline, or an SSO estate.
Who is it for?
security engineers and AI safety watchers