CVE-2024-58400
llama.cpp before b3701 contains an out-of-bounds read vulnerability in llama_decode_internal and llama_encode_internal that allows remote un
CVSS
8.2
High
EPSS
—
KEV
—
Exploit Today
—
0-100
Published: Oct 27, 2026 · Last modified: Oct 27, 2026 · CWE-129
Not enough EPSS history yet.
llama.cpp before b3701 contains an out-of-bounds read vulnerability in llama_decode_internal and llama_encode_internal that allows remote unauthenticated attackers to read memory by supplying unvalidated token IDs. Attackers can send integer-array prompts to the llama-server /completion endpoint with IDs outside 0 to n_vocab to read embedding memory out of bounds or crash the process.
- github.comhttps://github.com/ggml-org/llama.cpp/blob/b3681/examples/server/server.cpp#L789
- github.comhttps://github.com/ggml-org/llama.cpp/commit/eae597182cb61bbde0b26e7cec5999d28b9327fe
- github.comhttps://github.com/ggml-org/llama.cpp/commit/faf69d4237c9ae4d7f572b4674d1002463e8acd3
- github.comhttps://github.com/ggml-org/llama.cpp/pull/9359
- github.comhttps://github.com/ggml-org/llama.cpp/releases/tag/b3701
- www.vulncheck.comhttps://www.vulncheck.com/advisories/llama-cpp-before-b3701-out-of-bounds-read-in-llama-decode-via-token-ids
CVECVSSEPSSKEVRExploitTitleMod.
CVE-2026-1168167.8 HIG—
———llama.cpp b5899 before b10261 contains a heap out-of-bounds write vulnerability in llm_tokenizer_plamo2::build that allows attackers to corrupt heap memory via crafted GGUF models. Attackers can supply BYTE-typed token text like <0x-1> whose negative parsed value writes token ids before the bytes_ buffer during model loading, causing crashes or potential code execution.6hCVE-2026-1167316.1 MED—
———Furnace through 0.6.8.3 contains an out-of-bounds write vulnerability in DivEngine::loadFTM() that allows attackers to corrupt heap memory via an unchecked subsong index in the GROOVES block. Attackers can craft a FamiTracker module with a subsong count of 2 or more to write a groove pattern past the subsong vector, crashing Furnace when opened.9hCVE-2026-1165905.9 MED—
———snappy-java is a Java port of the snappy, a fast C++ compresser/decompresser developed by Google. Prior to 1.1.10.9, Snappy.arrayCopy(Object src, int offset, int byteLength, Object dest, int destOffset) passes caller-supplied ranges to native memcpy without validating either array, allowing negative or oversized byteLength and offset values to read or write outside array bounds and crash the JVM. SnappyOutputStream.rawWrite and SnappyInputStream.rawRead, including their typed-array write and read methods, pass caller-controlled ranges through this primitive, and typed-array byte-size calculations can overflow int. This issue is fixed in version 1.1.10.9.9hCVE-2026-116585——
———Go JOSE provides an implementation of the Javascript Object Signing and Encryption set of standards in Go, including support for JSON Web Encryption (JWE), JSON Web Signature (JWS), and JSON Web Token (JWT) standards. Prior to 4.1.5, decrypting untrusted JWE ciphertext with A128CBC-HS256, A192CBC-HS384, or A256CBC-HS512 can pass an empty ciphertext buffer to CBC-HMAC unpadding. Insufficient validation allows the empty buffer to be indexed, causing a panic and denial of service. Applications that decrypt attacker-controlled ciphertext with one of these content-encryption algorithms are affected. This issue is fixed in version 4.1.5.9hCVE-2026-1159166.5 MED—
———vLLM is an inference and serving engine for large language models. Prior to 0.32.0, vllm/tokenizers/hf.py computes max_token_id with an off-by-one upper bound, allowing a pre-tokenized prompt ID equal to the vocabulary size to pass validation. For a tensor-parallel size of one and an exact multiple-of-64 vocabulary, VocabParallelEmbedding.forward does not mask that value before F.embedding uses it, causing a CUDA device-side assertion and terminating the engine. This issue is fixed in version 0.32.0.13hCVE-2026-829477.1 HIG—
——0SOEM version 2.0.0 and earlier contains multiple vulnerabilities in frame processing where a malicious device can cause the master to read data outside allocated buffers. The software uses length and offset values from device messages without validating them against buffer sizes. This can cause the master to access invalid memory locations, potentially disclosing sensitive information or causing crashes.18h