Security1 hSeptember 21, 2026, 22:46

Hacktron details attack chain that compromised OpenAI employee accounts

Security researchers at Hacktron described a chain of vulnerabilities that compromised OpenAI employee accounts and reached internal GitHub repositories. The attack began with HEIF/HEIC image processing via ImageMagick and libheif, then moved through Discourse and OpenAI's SSO; Claude models assisted parts of the research.

DEV Community Open source

Hacktron details attack chain that compromised OpenAI employee accounts

Attack started with a HEIF/HEIC decoding flaw in libheif and ImageMagick
Chain moved through Discourse, OpenAI SSO and ChatGPT/Codex accounts
Final target was connected GitHub integrations and internal repositories
Anthropic's Claude models were used during the research

#OpenAI#Hacktron#GitHub#Anthropic

Hackers extract 1.6 million images and 27,000 videos from a single Flock camera

Security

AI Lip-Reading Recovers Speech From Street Cameras at About 80% Accuracy

Security

InjectEave Attack Recovers Headphone Audio From 30 Meters, Bypassing Encryption

Security

Hacktron details attack chain that compromised OpenAI employee accounts

Hackers extract 1.6 million images and 27,000 videos from a single Flock camera

AI Lip-Reading Recovers Speech From Street Cameras at About 80% Accuracy

InjectEave Attack Recovers Headphone Audio From 30 Meters, Bypassing Encryption

Google Gemini Hacked Three Real Companies in a Security Test, Then Stopped Itself