CVE-2025-49847: Buffer overflow in llama.cpp vocabulary loading
llama.cpp
A buffer overflow vulnerability in llama.cpp prior to version b5662 allows a malicious GGUF model vocabulary to trigger memory corruption via oversized token length casting. This can let an attacker overwrite memory beyond the intended buffer.
disclosed CVENo confirmed exploitation, so this record is tracked as a disclosed vulnerability and not counted as an incident. It moves into the counts if CISA or the vendor confirms exploitation. Why
- CVSS
- 8.8
- EPSS
- 0.01 · top 54% · 2026-11-13
- fix status
- Fix status unknown
- first report
- 17 Jun 2025
- last update
- 17 Jun 2025
- CVE ids
Timeline · 1
Sources · 1
Advisories and databases
- 2025-06-17CVE-2025-49847: llama.cpp is an inference of several LLM models in C/C++.NVD · nvd.nist.gov/vuln/detail/CVE-2025-49847
Titles and links belong to their publishers. The summary above is written by gcdRadar in its own words; data is CC BY-SA 4.0.
Classification
- lane
- Patch now
- AI role
- Attack on AI
- categories
- Infrastructure vulnerabilitySupply chain
- severity
- 2 · Disclosed vulnerability
- fix status
- Fix status unknown
- targets
- Framework / tooling, Open-weight model
- owasp llm top 10
- LLM03:2025 Supply Chain
- mitre atlas
- AML.T0010 AI Supply Chain Compromise
- dates
- happened 17 Jun 2025 · first reported 17 Jun 2025
- origin
- NVDVulnerability record from NVD / the GitHub Advisory Database. This product uses the NVD API but is not endorsed or certified by the NVD.
- record
- confidence 0.95 · created 2026-10-18 10:04Z · updated 2026-10-18 10:04Z
Lane, AI role, categories and severity come from the classifier; the OWASP and ATLAS rows are derived from the categories. How these are assigned
Copy a brief
This record as a short Markdown brief, for a ticket, a chat or a coding agent.
the brief as text
# CVE-2025-49847: Buffer overflow in llama.cpp vocabulary loading https://gcdradar.vercel.app/incidents/cve-2025-49847-buffer-overflow-in-llama-cpp-vocabulary-loadi · AIR-2025-0428 · closed 2025-06-17 - Lane: Patch now - Severity: 2 (Disclosed vulnerability) - Fix status: Fix status unknown - First reported: 2025-06-17 - AI role: Attack on AI - Products: llama.cpp - CVE ids: CVE-2025-49847 - CVSS: 8.8 ## Summary A buffer overflow vulnerability in llama.cpp prior to version b5662 allows a malicious GGUF model vocabulary to trigger memory corruption via oversized token length casting. This can let an attacker overwrite memory beyond the intended buffer. ## What to do No action recorded; read the sources below and the vendor's advisory. ## Latest updates - 2025-06-17: CVE-2025-49847: llama.cpp is an inference of several LLM models in C/C++. (https://nvd.nist.gov/vuln/detail/CVE-2025-49847) ## Sources - 2025-06-17 NVD: CVE-2025-49847: llama.cpp is an inference of several LLM models in C/C++. https://nvd.nist.gov/vuln/detail/CVE-2025-49847 Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.
Related
- closed7.3llama.cpp CVE-2026-78147: llama.cpp RPC server vulnerable to insecure deserialization
Patch now · Disclosed · since 23 Aug 2026 · CVE-2026-78147
AI - closed8.8llama.cpp Heap-based buffer overflow in llama.cpp header.n_kv
Patch now · Disclosed · since 26 Feb 2024 · CVE-2024-23605
AI - closed8.8llama.cpp Heap-based buffer overflow in llama.cpp info-ne
Patch now · Disclosed · since 26 Feb 2024 · CVE-2024-21802
AI - closed8.8llama.cpp Heap-based buffer overflow in llama.cpp gguf_fread_str
Patch now · Disclosed · since 26 Feb 2024 · CVE-2024-23496
AI - closed7.5vLLM CVE-2026-41523: vLLM Security Check Bypass via assert Statement
Patch now · Disclosed · since 16 Jun 2026 · updated 22 Jun 2026 · CVE-2026-41523
AI - closedllama.cpp Misconfigured Llama.cpp Servers Expose Sensitive User Prompts
Breaches & attacks · Exploited · since 11 Apr 2025
AI