ggml llama.cpp是ggml组织的一个在本地执行大语言模型推理的引擎。 ggml llama.cpp b3978版本至b9058版本存在安全漏洞,该漏洞源于DRY采样器存在整数下溢和越界读取问题,攻击者可通过向/v1/completions或/v1/chat/completions端点发送dry_allowed_length设置为INT32_MIN的特制HTTP请求,触发堆缓冲区下溢,可能导致服务器崩溃并拒绝服务,或通过读取内存垃圾值破坏token采样概率。
Although we use advanced large model technology, its output may still contain inaccurate or outdated information.Shenlong tries to ensure data accuracy, but please verify and judge based on the actual situation.
| # | POC Description | Source Link | Shenlong Link |
|---|
| CVE-2026-43632 | 8.1 HIGH | llama.cpp b7492–b9060 Use-After-Free in Tokenization Endpoints |
| CVE-2026-43631 | 8.1 HIGH | llama.cpp b7492–b9060 Use-After-Free RCE via llama-server |
| CVE-2026-43629 | 8.1 HIGH | llama.cpp b4882–b9058 Buffer Overflow in KV Cache State Restore |
| CVE-2026-43622 | 7.8 HIGH | llama.cpp b1886–b7445 Double Free via llama-android.cpp |
| CVE-2026-43627 | 7.8 HIGH | llama.cpp b1283–b9058 Integer Overflow in llama_batch_init() Function |
| CVE-2026-70638 | 7.8 HIGH | llama.cpp b1886–b7445 Integer Overflow via new_1batch() in llama-android.cpp |
| CVE-2026-70640 | 7.0 HIGH | llama.cpp b1886–b7445 Race Condition Use-After-Free via llama-android.cpp |
| CVE-2026-43630 | 6.5 MEDIUM | llama.cpp b5702–b7653 Out-of-Bounds Read Information Disclosure |
| CVE-2026-70639 | 5.5 MEDIUM | llama.cpp b1886–b7445 Null Pointer Dereference DoS via llama-android.cpp |
No comments yet