libheif EUVD-2026-83020

Requires libheif built with WITH_UNCOMPRESSED_CODEC enabled AND an application that uses the tiled writing API (heif_context_add_image_tile → ImageItem_uncompressed::add_image_tile → unc_encoder::encode_tile) to ENCODE, not decode. The attacker must be able to supply an independently constructed tile image whose component-plane width/height exceed the prototype tile geometry; the pre-fix encode_tile path lacked the check_component_sizes()/compatibility gate, so the mismatch is what triggers the overflow. Limiting factors: this is an encoder-side flaw (most libheif deployments only decode), the uncompressed codec is a build-time option that is not always compiled in, and it needs an application design that feeds untrusted image tiles into the tiled encoder - hence CVSS AC:H. No authentication or user interaction is required by the library itself (PR:N/UI:N), but reachability depends entirely on the host application's use of this specific API.

The vendor CVSS of 7.4 (CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:H/A:H) is defensible but the AC:H rating and the nature of the flaw materially cap real-world risk. Critically, this is an ENCODER-side heap out-of-bounds write, not the usual libheif decoder bug: it lives in unc_encoder::encode_tile()/unc_encoder_component_interleave::encode_tile() and is reached through the tiled-writing API heif_context_add_image_tile()/ImageItem_uncompressed::add_image_tile(). Exploitation therefore requires an application that (a) is built WITH_UNCOMPRESSED_CODEC and (b) drives libheif's tiled encoding path with attacker-influenced tile images whose component-plane dimensions differ from the prototype tile geometry - a far narrower and less common surface than simply decoding a malicious HEIF/AVIF. No EPSS score, KEV listing, or POC is provided in the input, and there is no SSVC decision data; the only public artifacts are the GitHub Security Advisory, the fix commit, and the v1.23.2 release. Exploitation status: no public exploit identified at time of analysis. The confidentiality impact is correctly N (this is a write primitive, not a leak), while I:H/A:H reflect heap corruption that can crash or potentially be steered. Net: a genuine memory-corruption bug in a widely-used library, but the encoder-only, build-flag-gated, high-complexity preconditions make it a moderate priority rather than a drop-everything issue for most consumers, who primarily use libheif to DECODE.

An application that uses libheif (built with WITH_UNCOMPRESSED_CODEC) to author tiled uncompressed HEIF images accepts image tiles derived from untrusted input and adds them via heif_context_add_image_tile(). An attacker supplies a tile whose component-plane dimensions are larger than the prototype tile geometry; encode_tile() sizes the output buffer from the configured geometry but copies the tile's actual (oversized) plane, writing attacker-controlled data past the heap output buffer. No public exploit code is identified at time of analysis.