CVE-2026-107183
8.1 HIGHpublic exploit availablePublished 2026-10-07 · Updated 2026-10-07
AI risk analysis
- Summary
- The vulnerability is a use-after-free and double free in llama.cpp's chat parser that allows remote attackers to corrupt heap memory by submitting crafted tool-id tags after tool-close tags, potentially leading to arbitrary code execution or denial of service.
- Exploitability
- Exploitation requires sending a specially crafted POST /completion request with a chat_parser containing a tool-id after a tool-close tag, making it moderately difficult but possible for unauthenticated attackers to trigger heap corruption.
- Blast radius
- If exploited, this vulnerability could affect any system running llama-server with the affected version of llama.cpp, potentially allowing remote code execution or complete service denial across all chat-based applications using this library.
- Detection
- No reliable host or network indicator is derivable from the published description.
- Prioritized remediation
- Upgrade to llama.cpp version b11393 or later to fix the use-after-free and double free vulnerability in common_chat_peg_mapper::map.
Analysis generated locally by garnet-coder (no data left the box). AI-assisted — verify against primary sources before acting.
NVD description
llama.cpp before b11393 contains a use-after-free and double free vulnerability in common_chat_peg_mapper::map that allows unauthenticated remote attackers to corrupt heap memory via a dangling current_tool pointer. Attackers can submit a chat_parser in a POST /completion request emitting a tool-id after a tool-close tag to crash llama-server and shape a heap write primitive.
CVSS vector
CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H
Weaknesses
CWE-416
Public exploit & PoC references
- https://github.com/ggml-org/llama.cpp
- https://github.com/ggml-org/llama.cpp/blob/bed0a856606ee4a24a164066f73d2379447033f5/common/chat-peg-parser.cpp#L343-L355
- https://github.com/ggml-org/llama.cpp/blob/bed0a856606ee4a24a164066f73d2379447033f5/tools/server/server-schema.cpp#L321-L324
- https://github.com/ggml-org/llama.cpp/commit/dbe4c3ed42343f0a7ba0fd7e808ffeaca404d29f
- https://github.com/ggml-org/llama.cpp/pull/29942
All references
- https://github.com/ggml-org/llama.cpp
- https://github.com/ggml-org/llama.cpp/blob/bed0a856606ee4a24a164066f73d2379447033f5/common/chat-peg-parser.cpp#L343-L355
- https://github.com/ggml-org/llama.cpp/blob/bed0a856606ee4a24a164066f73d2379447033f5/tools/server/server-schema.cpp#L321-L324
- https://github.com/ggml-org/llama.cpp/commit/dbe4c3ed42343f0a7ba0fd7e808ffeaca404d29f
- https://github.com/ggml-org/llama.cpp/pull/29942
- https://www.vulncheck.com/advisories/llama-cpp-before-b11393-use-after-free-via-common-chat-peg-mapper-chat-parser
Source data: NVD (nvd.nist.gov), public domain. Exploit-DB.ai adds local AI analysis for defensive use only.
Related CVEs
- HIGHCVE-2026-100209
- HIGHCVE-2026-100211
- CRITICALCVE-2026-100219
- HIGHCVE-2026-100805
- CRITICALCVE-2026-109707
- HIGHCVE-2026-110148
- HIGHCVE-2026-20531
- CRITICALCVE-2026-6928
Related by shared AI tags and CWE weakness class. Browse the full archive.