CVE-2026-32814

Vulnerability

In libheif versions 1.21.2 and prior, when decoding a HEIF grid image with strict_decoding=false (the default), a corrupted tile silently fails to decode and the library returns heif_error_Ok with no indication of failure [1]. The canvas is allocated via create_clone_image_at_new_size() → plane.alloc() → new (std::nothrow) uint8_t[allocation_size] which does not zero the memory; only the alpha plane is explicitly initialized via fill_plane(), so the Y, Cb, and Cr planes retain whatever heap data was previously at that address [1]. The failed tile's region of the canvas is never written, delivering uninitialized heap data (4,096 bytes per Y/Cb/Cr plane, totaling 12,288+ bytes) to the caller as decoded pixel values [1]. Affected versions: all up to and including 1.21.2.

Exploitation

An attacker with the ability to deliver a crafted .heic or .avif file to an application using libheif can trigger the leak [1]. No authentication or special network position is required if the file is uploaded to a server that decodes it. The attacker constructs a HEIF/AVIF grid image where at least one tile is deliberately corrupted. When the application calls heif_decode_image() with default settings (strict_decoding=false), the corrupt tile fails to decode silently, the function returns heif_error_Ok, and the uninitialized heap region becomes part of the decoded image output [1]. The calling application has no indication of failure and may re-encode the result (e.g., as PNG/JPEG for thumbnails, CDN, or social media) [1].

Impact

Successful exploitation leaks the contents of heap memory from the server process into the decoded image, potentially exposing sensitive cross-user data such as authentication tokens, session identifiers, database query results, or other users' image data [1]. In server-side image processing pipelines, this information can be transmitted to the attacker or embedded in downstream assets. The CIA outcome is primarily information disclosure; no code execution or privilege escalation is achieved, but the leaked heap memory can reveal secrets that facilitate further attacks [1].

Mitigation

This issue has been fixed in libheif version 1.22.0 [1][2]. Users should upgrade to 1.22.0 or later. There is no known workaround for unpatched versions other than enabling strict_decoding=true, which causes the entire decode to fail on corrupt tiles and may break legitimate files; the default remains false for backward compatibility [1]. No CVE entry for the KEV list was provided.