cgltf CVE-2026-112018

Exploitation requires that an application using cgltf (jkuhlmann/cgltf, versions through 1.15) parse an attacker-supplied glTF or GLB file AND NOT call cgltf_validate() before unpacking accessor data - validation is the explicit mitigating control that would reject the out-of-bounds sparse indices. The specific trigger is a sparse accessor whose indices exceed the output buffer, reached through cgltf_accessor_unpack_floats(). Limiting factors: the attack is local (the crafted file must be delivered to and opened/loaded by the victim, hence user interaction per UI:P), and any integration that invokes cgltf_validate() on untrusted input is not vulnerable. No network service exposure or authentication is involved; risk is confined to toolchains, engines, viewers, or importers that ingest untrusted 3D assets without validation.

This is a genuine memory-corruption flaw (CWE-787 heap out-of-bounds write) in cgltf, a widely-used single-header C glTF/GLB loader, but the real-world risk is moderate and heavily conditional rather than critical. The provided CVSS 4.0 vector (AV:L/AC:L/AT:P/PR:N/UI:P/VC:H/VI:H/VA:H) scores 7.3 and tells a consistent story: exploitation is LOCAL (the attacker must get a crafted asset to the victim), requires passive user interaction (UI:P - loading/opening a model file), and carries an attack requirement (AT:P) because the condition 'cgltf_validate() not called' must hold. High impact to confidentiality/integrity/availability (VC/VI/VA:H) reflects the potential for code execution via heap corruption, but 'potentially execute code' is the vendor/researcher framing - reliable RCE from a heap OOB write is not demonstrated in the supplied data. No EPSS score, no CISA KEV listing, and no weaponized exploit are present in the input; the referenced HackMD writeup is a technical root-cause analysis rather than confirmed exploit code. Exploitation status is therefore: no public exploit identified at time of analysis. Net: a legitimate issue to patch for anyone embedding cgltf in a pipeline that ingests untrusted models, but not a drop-everything, mass-exploitation priority - the local vector, required user interaction, and the strong mitigating factor that calling cgltf_validate() prevents the condition all cap the realistic risk.

An attacker crafts a malicious glTF or GLB 3D model file containing a sparse accessor with oversized/unchecked indices and delivers it to a victim (e.g., via a model-sharing site, game mod, asset marketplace, or email). When a victim application built on cgltf loads the file without first calling cgltf_validate(), cgltf_accessor_unpack_floats() writes past the end of the output heap buffer, corrupting memory and crashing the process, with potential for code execution. No public, weaponized exploit was identified in the provided intelligence; only a technical root-cause writeup is referenced.