Researchers use AI to find widespread software decoder flaw 

Researchers disclosed “HEIF Heist,” a memory-corruption vulnerability in common HEIF/HEIC/AVIF decoders (libheif/libde265) that can leak in-memory data and lead to remote code execution; the team used AI models to accelerate exploit development and demonstrated a proof-of-concept compromise of OpenAI employee accounts and access to internal repositories, while noting patches exist but unpatched deployments remain at risk.