cgltf EUVD-2026-103472

Exploitation requires the target application to parse an attacker-controlled glTF or GLB file using cgltf (version 1.15 or earlier) that invokes cgltf_accessor_unpack_floats() on a sparse accessor; the crafted file must define a sparse accessor whose base bufferView specifies a byteStride, which the code incorrectly uses to advance the sparse values cursor past the allocated buffer. The flaw is reachable even on files that pass cgltf_validate(), so input validation at that layer does not block it. Limiting factors: the vector is local and file-based (UI:P) - a user or automated pipeline must actively load the malicious asset; there is no remote or network-triggered path, and no code execution, so realistic impact is limited to application crashes and limited heap information disclosure into vertex data.

This is a moderate-risk memory-safety defect, not a top-tier emergency. The vendor-supplied CVSS 4.0 vector (AV:L/AC:L/AT:N/PR:N/UI:P/VC:L/VI:N/VA:H) correctly frames it as a local, file-driven issue: exploitation requires a victim to process an attacker-supplied glTF/GLB file (UI:P), so there is no remote, unauthenticated, network-facing path despite the serious-sounding 'heap out-of-bounds read.' The primary impact is availability (VA:H - crash/DoS of the consuming application) with only limited confidentiality exposure (VC:L - adjacent heap bytes leaked into vertex data), and no integrity impact. No EPSS score was provided, and the CVE is not listed in CISA KEV, so there is no evidence of active or widespread exploitation. A detailed public technical write-up (hackmd) and a VulnCheck advisory exist with exact source-line references, which lowers the barrier to building a trigger even though no packaged exploit code is explicitly indicated in the input. Signals are largely consistent: a low/local-vector CVSS, absence of KEV, and absence of EPSS all point to this being a real but defender-manageable bug whose true risk depends entirely on whether an application feeds untrusted 3D assets to cgltf. Note: the input conflates CWE-125 (out-of-bounds READ) with the 'Buffer Overflow' tag - the actual flaw is an over-read, not a write, which caps the realistic worst case at information disclosure and crashes rather than code execution.

An attacker crafts a malicious glTF or GLB model that passes cgltf_validate() but declares a sparse accessor whose values cursor is advanced by the base bufferView byteStride, causing cgltf_accessor_unpack_floats() to read past the heap allocation. A victim opens or imports this model into a game engine, asset pipeline, or 3D viewer that embeds cgltf, leaking adjacent heap memory into parsed vertex data or crashing the application. A public technical write-up documenting the exact vulnerable code paths exists, lowering the effort to reproduce; no packaged weaponized exploit is confirmed in the provided data.