PatchSiren

PatchSiren cyber security CVE debrief

CVE-2026-84286 Turboderp-org CVE debrief

A vulnerability in ExLlamaV3's EXL3 loader allows an out-of-bounds read and write via a negative array index. A malicious model can cause a reproducible crash and potentially achieve controlled indirect kernel dispatch during model load under specific conditions. The vulnerability is triggered by an untrusted, model-supplied index used to select a CUDA kernel without bounds verification. This issue can lead to a crash and, under certain host memory layouts, may allow for controlled indirect kernel dispatch. Defenders should assess exposure and prioritize verification of model inputs and kernel indices.

Vendor
Turboderp-org
Product
ExLlamaV3
CVSS
Unknown
CISA KEV
Not listed in stored evidence
Original CVE published
2026-10-12
Original CVE updated
2026-10-12
Advisory published
2026-10-12
Advisory updated
2026-10-12

Who should care

Defenders responsible for deploying and managing ExLlamaV3 models should assess exposure and prioritize verification of model inputs and kernel indices. This includes reviewing the official advisory or CVE record to validate affected scope, severity, and vendor guidance. Additionally, defenders should plan for vendor-supported updates or mitigations through normal change control where exposure is confirmed and review compensating controls for exposed and

Why it matters

Defenders should prioritize verifying model inputs and validating kernel indices to prevent potential exploitation of ExLlamaV3's EXL3 loader vulnerability.

  • Potential for controlled indirect kernel dispatch during model load
  • Reproducible crash via malicious model
  • Need for input validation and bounds checking

Technical summary

The EXL3 loader in ExLlamaV3 uses an untrusted, model-supplied index to select a CUDA kernel without bounds verification, allowing for out-of-bounds reads and writes. This vulnerability can lead to a reproducible crash and, under specific conditions involving certain host memory layouts, may allow for controlled indirect kernel dispatch during model load. The issue arises from the lack of input validation and bounds checking in the kernel index selection process. Defenders should prioritize verifying model inputs and validating kernel indices to prevent potential exploitation.

Defensive priority

Defenders should prioritize verifying model inputs and validating kernel indices to prevent potential exploitation.

Recommended defensive actions

  • Verify model inputs and validate kernel indices
  • Implement bounds checking for kernel table indices
  • Monitor for suspicious model loading activity
  • Review compensating controls for exposed systems while remediation is scheduled and verified
  • Check relevant monitoring, detection, and logs for exposed assets that need extra review
  • Confirm whether affected product deployments exist in managed environments and assign an owner for follow-up
  • Plan vendor-supported updates or mitigations through normal change control where exposure is confirmed

Evidence notes

The CVE record and source item provide details on the vulnerability, but additional information on affected versions and remediation is limited. Further verification is needed to determine the full scope of affected systems and to develop effective mitigations. Defenders should review the official advisory and CVE record for more information and track exceptions, retest remediated assets, and close the item only after evidence is documented.

Sources and references

Verified primary and authoritative sources

  • CVE-2026-84286 CVE Program record

    Publisher, destination, and source semantics verified

    URL: https://www.cve.org/CVERecord?id=CVE-2026-84286

    CVE Program - Official CVE Program record with source-provided CVE metadata.

  • CVE-2026-84286 NVD vulnerability detail

    Publisher, destination, and source semantics verified

    URL: https://nvd.nist.gov/vuln/detail/CVE-2026-84286

    NIST National Vulnerability Database - Official NIST NVD detail page and source-specific vulnerability assessment.

Supplemental references

Methodology and review provenance

AI-assisted synthesis based on stored public vulnerability evidence. System validation, approval state, and publication status do not by themselves establish human review of this revision. PatchSiren helps prioritize defensive review and does not prove exposure or remediation on any system.