Read the docs
TI-2025-32661 Aliases CVE-2025-52566

llama.cpp tokenizer signed vs. unsigned heap overflow

Priority HIGH 73CWE-119Published · Updated

Key facts

Attack vector
Network
Privileges required
None
User interaction
Required
Attack complexity
Low
Impact type
Code execution
Confidentiality · integrity · availability
High · High · High
Fix status
No fix listed

Attack profile from CVSS 3.1 · NVD. Fix status from CNA. Impact type from CWE-119 and the CVSS impact.

Summary

llama.cpp is an inference of several LLM models in C/C++. Prior to version b5721, there is a signed vs. unsigned integer overflow in llama.cpp's tokenizer implementation (llama_vocab::tokenize) (src/llama-vocab.cpp:3036) resulting in unintended behavior in tokens copying size comparison. Allowing heap-overflowing llama.cpp inferencing engine with carefully manipulated text input during tokenization process. This issue has been patched in version b5721.

Why it matters now

Exploitation Proof of concept CISA SSVC · not in CISA KEV
EPSS 0.00363 30th percentileScored 2026-11-07
Public exploit Yes 1 exploit referenceTagged by NVD
Fix No fix listed No fixed version in the sources yet

SSVC decision points

CISA ADP · SSVC 2.0.3

Exploitation
Poc
Automatable
No
Technical impact
Total

The decision (Track, Attend, Act) also depends on your mission and well-being impact; see CISA's SSVC guide.

Affected products

VendorProductAffected versionsFixed inStatusSource
ggml-orgllama.cpp< b5721—No fix listedCNA
  • llama.cpp Vendorggml-org Affected < b5721 Fixed in — No fix listed CNA

CPE configurations NVD · 1

CPEVersionsRoleVendor
cpe:2.3:a:ggml:llama.cpp:*:*:*:*:*:*:*:*before b5721Vulnerableggml-org
  • cpe:2.3:a:ggml:llama.cpp:*:*:*:*:*:*:*:* Versions before b5721 Vulnerable ·ggml-org

Remediation

No fixed version listed by the sources yet.

Vendor assessments

Red Hat

Red Hat's page for this CVE →

Not affected: 1 product
  • Red Hat

Timeline

  1. 2026-10-02 Threatist record created from CVE-2025-52566

References

Exploit 1

Patch 1

  • TI-2026-26459 llama.cpp: Unauthenticated RCE via GRAPH_COMPUTE buffer=0 bypass in llama.cpp RPC backend CVSS 3.1CRITICAL 9.8
  • TI-2024-28446 llama.cpp allows Arbitrary Address Read in rpc_server::get_tensor CVSS 3.1CRITICAL 9.8
  • TI-2026-17570 llama.cpp has Out-of-bounds Write in llama-server CVSS 3.1CRITICAL 9.8
  • TI-2024-28447 llama.cpp allows write-what-where in rpc_server::set_tensor CVSS 3.1CRITICAL 10.0
  • TI-2024-15627 Integer overflow in ggml-org llama.cpp CVSS 3.1CRITICAL 9.8
  • TI-2024-14482 Heap-based buffer overflow in ggml-org llama.cpp CVSS 3.1CRITICAL 9.8
See all 28 →

Sources

This product uses the NVD API but is not endorsed or certified by the NVD. Sources and licences