CVE-2025-52566

What it is

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.

At a glance

Exploitability

Can be exploited locally on the machine without logging in, but needs a user to act. High risk of data exposure, data tampering, service disruption.

Exploit odds

0.4% chance of exploitation in the next 30 days (EPSS).

Fix

A patch, advisory or fixed version is referenced. Upgrade or apply it.

Timeline

1. 2025-06-24 04:15 UTC PublishedDisclosed publicly on CVE.org and NVD
2. 2026-06-17 09:36 UTC Last updatedMost recent change to the record

Why it ranks here

• Exploited in the wild: no, CISA does not list it.
• Exploit odds: 0.4% in the next 30 days (EPSS).
• Severity: 8.6 HIGH (CVSS).
• Fix: a patch or advisory is linked.

Details

Published

2025-06-24 04:15 UTC

Last modified

2026-06-17 09:36 UTC

Exploit probability

0.4% · EPSS, next 30 days

Weakness

CWE-119 · Buffer errors

Record status

Analyzed

Affected

• Ggml llama.cpp < b5721

Attack profile CVSS 3.1 · 8.6

Attack vector

Local

Complexity

Low

Privileges needed

None

User interaction

Required

Scope

Changed

Confidentiality

High

Integrity

High

Availability

High

Is it in your code?

Hand this to your coding agent. It searches your project for the affected products and checks versions. Nothing leaves your machine.

Read https://cve-list.com/agent/cve/CVE-2025-52566.md and follow it. Check whether this project is affected by CVE-2025-52566. Read only. Don't change anything and don't upload my code. Tell me yes, no or not sure, with the file as proof.

Copy as ticket · More options · Status badge