CVE-2026-43631 ggml-org CVE debrief

Who should care

Users and administrators of llama.cpp, especially those using affected versions b7492 through b9060, should be aware of this critical vulnerability and take immediate action to mitigate the risk. Affected operator and platform impact should be carefully reviewed, and security teams should prioritize remediation efforts.

Technical summary

The CVE-2026-43631 vulnerability exists in llama.cpp builds b7492 through b9060, specifically in the vocab pointer of llama-server when the --sleep-idle-seconds feature is enabled. This allows unauthenticated remote attackers to execute arbitrary code by sending requests to affected endpoints while the server transitions to sleep mode, causing concurrent worker threads to dereference a freed vocab pointer that can be reclaimed with attacker-controlled data. Affected product context indicates that users and administrators of llama.cpp, especially those using affected versions, should be aware of this critical vulnerability and take immediate action to mitigate the risk.

Defensive priority

Critical vulnerability in llama.cpp, allowing unauthenticated remote code execution with high CVSS score of 9.2.

Recommended defensive actions

• Confirm whether affected product deployments exist in managed environments and assign an owner for follow-up.
• Review the supplied official advisory or CVE record to validate affected scope, severity, and vendor guidance.
• Plan vendor-supported updates or mitigations through normal change control where exposure is confirmed.
• 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.
• Track exceptions, retest remediated assets, and close the item only after evidence is documented.
• Perform regular security audits and vulnerability assessments.

Evidence notes

The CVE-2026-43631 vulnerability exists in llama.cpp builds b7492 through b9060, specifically in the vocab pointer of llama-server when the --sleep-idle-seconds feature is enabled. This allows unauthenticated remote attackers to execute arbitrary code by sending requests to affected endpoints while the server transitions to sleep mode, causing concurrent worker threads to dereference a freed vocab pointer that can be reclaimed with attacker-controlled data.

Official resources