Skip to content- Sources: primary
- Summary: The video traces a libheif bug in heif_pixel_image overlay code that HackerOne researchers used against a Discourse forum to reach OpenAI internal source code. Negative overlay offsets produce a wrong width, and the alpha blending path computes the output pointer without subtracting the x start offset, so writes spill into following rows and past the heap buffer. The fixing commit had a benign message and no CVE.
- Why it matters: Services that decode user-uploaded HEIC images with libheif should confirm they run a build that includes the overlay fix, since scanners keyed on CVEs will miss it.