llama.cpp: Heap out-of-bounds write in plamo2 tokenizer via crafted GGUF model
Summary
llama.cpp versions from b5899 up to but not including b10261 contain a heap out-of-bounds write in the plamo2 tokenizer during model loading. A crafted GGUF model file with a negative byte token value, such as <0x-1>, can overwrite heap memory before the bytes_ buffer. An attacker who gets a user or service to load such a model can crash the application or possibly execute code.
Affected
- llama.cpp b5899 up to but not including b10261
- Downstream builds, bundled copies and Python wrappers may include affected versions; check each product's version.
Fixed in
- llama.cpp b10261 or later (fix: commit cf06ad7, PR #26511)
Workaround
- No vendor workaround is documented. Update to b10261 or later.
- Until the update is applied, load GGUF model files only from trusted sources.
Exploitation
- Exploitation in the wild: not known.
- Public proof of concept: none found for this CVE.
- CISA SSVC: exploitation none, automatable no, technical impact total.
Our recommendation
- Update llama.cpp to b10261 or later, including distribution packages, container images and Python wrappers.
- Load only GGUF models from trusted sources until the update is in place.
- Check where llama.cpp runs (developer workstations, internal AI services, automated model pulls) and prioritise those systems.
- Monitor for crashes of llama.cpp processes during model loading.
Public exploits
PoC-in-GitHub, Metasploit and Exploit-DB list nothing for this CVE.
Research sources
- CVE record advisory (VulnCheck) www.vulncheck.com
- llama.cpp release b10261 github.com
- Fix commit cf06ad7 (vocab: validate plamo2 byte tokens) github.com
- Fix pull request #26511 github.com
- Vulnerable code at b10259 (llama-vocab.cpp) github.com
- Ubuntu security tracker CVE-2026-116816 ubuntu.com
Researched and written up in English and German by an AI (Claude by Anthropic) based on NVD, CISA, EPSS and the sources listed. The vendor information is authoritative. Researched on 1 Nov 2026, 10:03.
Original description (NVD)
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.
NVD references (6)
- https://github.com/ggml-org/llama.cpp/blob/b10259/src/llama-vocab.cpp#L1376-L1377
- https://github.com/ggml-org/llama.cpp/commit/cf06ad7dfe79e7bf2934150d4fefe1ce8db05490
- https://github.com/ggml-org/llama.cpp/pull/25969
- https://github.com/ggml-org/llama.cpp/pull/26511
- https://github.com/ggml-org/llama.cpp/releases/tag/b10261
- https://www.vulncheck.com/advisories/llama-cpp-b5899-before-b10261-heap-out-of-bounds-write-via-plamo2-tokeni…
Sources
The tracker combines official, freely usable data sources – updated continuously:
- NVD (NIST) – CVEs, rating (CVSS), affected products
- CISA KEV – vulnerabilities confirmed by CISA as actively exploited
- CISA Vulnrichment – CISA assessment (SSVC) in the official CVE record
- FIRST EPSS – probability of exploitation
- Microsoft MSRC – Patch Tuesday: vulnerabilities, severity, updates
- PoC-in-GitHub – public proof-of-concept repositories
- Metasploit (Rapid7) – modules of the Metasploit Framework
- Exploit-DB (OffSec) – archive of public exploits (links only)
- Patch days: official vendor pages (linked per vendor) (Vendor patch days)
This product uses data from the NVD API but is not endorsed or certified by the NVD. EPSS data © FIRST.org.
Affected? We can help.
We check whether your systems are affected, install updates and close the gap – continuously with the System Health Checker if you wish.