Executive brief
llama.cpp is a C/C++ library for running large language models locally. The Jinja template parser component contains a flaw that can be triggered with specific input, causing the application to crash due to a failed assertion. An attacker with local access could exploit this to cause a denial of service.
Technical details
A reachable assertion vulnerability exists in the Jinja Minja Template Parser component of llama.cpp (commit e15efe0), specifically in the file common/jinja/parser.cpp. The vulnerability is triggered by crafted template input such as {{9|9|{ which causes an assertion to fail. The attack vector requires local access to the system running llama.cpp. The exploit has been publicly disclosed, and the vendor has been informed but has not yet responded with a patch or fix.
Affected products
- ggml-org llama.cpp commit e15efe0 and potentially others
Timeline
- 2026-08-03: disclosed: Publicly disclosed vulnerability
- 2026-08-03: other: Vendor informed but has not responded