CVE-2026-43628: CWE-191 Integer Underflow (Wrap or Wraparound) in ggml-org llama.cpp
An integer underflow and out-of-bounds read vulnerability exists in llama.cpp versions b3978 and 0.11.0 in the DRY sampler component. This flaw allows unauthenticated attackers to trigger a heap buffer underflow by sending a crafted HTTP request with dry_allowed_length set to INT32_MIN to specific API endpoints. Exploitation can cause the server to crash with a segmentation fault, resulting in denial of service, or corrupt token sampling probabilities by reading invalid memory values.
AI Analysis
Technical Summary
llama.cpp builds b3978 through 0.11.0 contain an integer underflow and out-of-bounds read vulnerability in the DRY sampler. An attacker can send a crafted HTTP request with dry_allowed_length set to INT32_MIN to the /v1/completions or /v1/chat/completions endpoints, triggering a heap buffer underflow. This leads to a segmentation fault (SIGSEGV) crash causing denial of service or corruption of token sampling probabilities due to reading garbage memory values before the allocated buffer.
Potential Impact
Successful exploitation results in denial of service by crashing the server process with a segmentation fault. Additionally, it can cause corruption of token sampling probabilities, potentially affecting the integrity of model outputs. No evidence of privilege escalation or remote code execution is indicated.
Mitigation Recommendations
Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. No official fix or temporary workaround is currently documented. Until a patch is available, restrict access to the affected endpoints or implement input validation to prevent dry_allowed_length from being set to INT32_MIN.
CVE-2026-43628: CWE-191 Integer Underflow (Wrap or Wraparound) in ggml-org llama.cpp
Description
An integer underflow and out-of-bounds read vulnerability exists in llama.cpp versions b3978 and 0.11.0 in the DRY sampler component. This flaw allows unauthenticated attackers to trigger a heap buffer underflow by sending a crafted HTTP request with dry_allowed_length set to INT32_MIN to specific API endpoints. Exploitation can cause the server to crash with a segmentation fault, resulting in denial of service, or corrupt token sampling probabilities by reading invalid memory values.
CVSS v4.0
Score 8.5high
Affected software
pkg:github/ggml-org/llama.cppRun on your own infrastructure? Check whether these packages are installed with threat-finder — our free open-source scanner.
Weaknesses
AI-Powered Analysis
Machine-generated threat intelligence
Technical Analysis
llama.cpp builds b3978 through 0.11.0 contain an integer underflow and out-of-bounds read vulnerability in the DRY sampler. An attacker can send a crafted HTTP request with dry_allowed_length set to INT32_MIN to the /v1/completions or /v1/chat/completions endpoints, triggering a heap buffer underflow. This leads to a segmentation fault (SIGSEGV) crash causing denial of service or corruption of token sampling probabilities due to reading garbage memory values before the allocated buffer.
Potential Impact
Successful exploitation results in denial of service by crashing the server process with a segmentation fault. Additionally, it can cause corruption of token sampling probabilities, potentially affecting the integrity of model outputs. No evidence of privilege escalation or remote code execution is indicated.
Mitigation Recommendations
Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. No official fix or temporary workaround is currently documented. Until a patch is available, restrict access to the affected endpoints or implement input validation to prevent dry_allowed_length from being set to INT32_MIN.
Technical Details
- Data Version
- 5.2
- Assigner Short Name
- VulnCheck
- Date Reserved
- 2026-05-01T18:22:45.641Z
- Cvss Version
- 4.0
- State
- PUBLISHED
- Remediation Level
- null
Threat ID: 6a750702bf8831d5395f59bf
Added to database: 08/06/2026, 22:13:22 UTC
Last enriched: 08/06/2026, 23:11:42 UTC
Last updated: 08/07/2026, 00:01:03 UTC
Views: 2
Community Reviews
0 reviewsCrowdsource mitigation strategies, share intel context, and vote on the most helpful responses. Sign in to add your voice and help keep defenders ahead.
Want to contribute mitigation steps or threat intel context? Sign in or create an account to join the community discussion.
Actions
Updates to AI analysis require Pro Console access. Upgrade inside Console → Billing.
External Links
Need more coverage?
Upgrade to Pro Console for AI refresh and higher limits.
For incident response and remediation, OffSeq services can help resolve threats faster.
Latest Threats
Check if your credentials are on the dark web
Instant breach scanning across billions of leaked records. Free tier available.