ExLlamaV3: Out-of-bounds memory access in EXL3 model loader
Summary
ExLlamaV3, a library for running large language models locally on GPUs, selects CUDA kernels using an index taken from the model file without checking that it is in range. A malicious EXL3 model can trigger out-of-range function pointer access, which reliably crashes the process. In certain host memory layouts, an attacker may be able to achieve controlled indirect kernel dispatch during model loading.
Affected
- ExLlamaV3 versions containing the EXL3 loader (exllamav3_ext reconstruct kernels)
- Exact affected version range: Unknown. The CVE record gives only the placeholder product version "3".
Fixed in
- Fix: GitHub pull request #310 (validates the kernel index before dispatch), merged into the dev branch on 2026-09-02
- First released version containing the fix: Unknown.
Workaround
- Load EXL3 models only from trusted sources and verify file hashes before use.
- Load models only in an isolated environment, such as a container or VM without access to sensitive data or the internal network.
- Do not accept model uploads from untrusted users on shared inference servers.
Exploitation
- CISA SSVC: Exploitation none, Automatable yes, Technical Impact total.
- No evidence of exploitation in the wild; not listed in CISA KEV.
- No public proof of concept found. The Positive Technologies record describes a crash-only denial of service.
Our recommendation
- Inventory all ExLlamaV3 installations on workstations and GPU servers.
- Restrict EXL3 model sources to trusted providers and verify hashes.
- Isolate model loading from sensitive systems.
- Update ExLlamaV3 as soon as a released version includes PR #310; monitor VU#369611 and the repository releases until then.
Public exploits
PoC-in-GitHub, Metasploit and Exploit-DB list nothing for this CVE.
Research sources
- GitHub Pull Request #310: Validate reconstruct kernel indices before dispatch github.com
- CERT/CC Vulnerability Note VU#369611 kb.cert.org
- CVE-2026-84286 on OpenCVE app.opencve.io
- CVE-2026-84286 on Vulners vulners.com
- Positive Technologies PT-2026-90534 on Vulners vulners.com
- ExLlamaV3 issues github.com
Researched and written up in English and German by an AI (Claude by Anthropic) based on NVD, CISA, EPSS and the sources listed. The vendor information is authoritative. Researched on 30 Oct 2026, 19:13.
Original description (NVD)
ExLlamaV3’s EXL3 loader uses an untrusted, model-supplied index to select a CUDA kernel without first verifying that the index is within bounds. A malicious model can force the code to read and invoke an out‑of‑range function pointer, resulting in a reproducible crash. Under specific conditions involving certain host memory layouts, it may be possible for an attacker to exploit this vulnerability to achieve controlled indirect kernel dispatch during model load.
Sources
The tracker combines official, freely usable data sources – updated continuously:
- NVD (NIST) – CVEs, rating (CVSS), affected products
- CISA KEV – vulnerabilities confirmed by CISA as actively exploited
- CISA Vulnrichment – CISA assessment (SSVC) in the official CVE record
- FIRST EPSS – probability of exploitation
- Microsoft MSRC – Patch Tuesday: vulnerabilities, severity, updates
- PoC-in-GitHub – public proof-of-concept repositories
- Metasploit (Rapid7) – modules of the Metasploit Framework
- Exploit-DB (OffSec) – archive of public exploits (links only)
- Patch days: official vendor pages (linked per vendor) (Vendor patch days)
This product uses data from the NVD API but is not endorsed or certified by the NVD. EPSS data © FIRST.org.
Affected? We can help.
We check whether your systems are affected, install updates and close the gap – continuously with the System Health Checker if you wish.