CVE-2024-23496

CVE-2024-23496 affects llama.cpp’s GGUF file parsing in the gguf_fread_str function. The root cause is an integer overflow when reading a string length p->n, leading to p->data = calloc(p->n + 1, 1) and a potentially undersized allocation. A heap-based buffer overflow can occur when the ...

CVE-2024-21825

Talos details CVE-2024-21825 in llama.cpp (GGUF library) focusing on parsing of GGUF_TYPE_ARRAY/GGUF_TYPE_STRING within gguf_init_from_file. An attacker-provided .gguf file can trigger a heap-based buffer overflow when kv->value.arr.n is large, due to an integer overflow in the allocation kv-&...

CVE-2024-23605

Affected: llama.cpp with GGUF header.n_kv parsing (commit 18c2e17). TALOS details a heap-based buffer overflow in gguf_init_from_file caused by allocating ctx->kv with header.n_kv, risking overflow when writing kv->key strings if header.n_kv overflows the allocation. Result: potential code ...

CVE-2024-21836

CVE-2024-21836 affects llama.cpp (GGUF library header.n_tensors parsing). Talos reports a heap-based buffer overflow caused by an integer overflow when allocating ctx->infos for a crafted .gguf file, which can lead to a heap overflow during reading tensor names (info->name). Vulnerable comp...

CVE-2024-21802

Talos reports a heap-based buffer overflow in llama.cpp (Commit 18c2e17) related to the GGUF library info->ne parsing when processing .gguf files, which can lead to remote code execution. The vulnerability stems from writing beyond a 4-element info->ne array during gguf_init_from_file, with...

CVE-2024-42478

CVE-2024-42478 concerns llama.cpp where an unsafe data pointer in the rpc_tensor structure can lead to arbitrary address reads. Several sources concur this is a code-level issue; the public CVE description states the vulnerability is fixed in b3561. Some connected advisories also advise upgrading...

CVE-2026-21869

CVE-2026-21869 affects llama.cpp prior to commit 55d4206c9, where the server’s completion endpoints parse the non‑negative constraint for the JSON input parameter n_discard without validation. A negative n_discard can cause a reversed range/offset in llama_memory_seq_rm/add, leading to out‑of‑bou...

CVE-2026-34159

The CVE-2026-34159 entry for llama.cpp describes an unauthenticated RCE via the RPC backend: prior to v.b8492, deserialize_tensor() omits bounds validation when tensor.buffer == 0, enabling an attacker to read/write arbitrary process memory through crafted GRAPH_COMPUTE messages. Combined with AL...