llama.cpp builds b1886 through b7445 contain a race condition use-after-free vulnerability in the LLaMA-Android JNI wrapper where bench_1model() and free_1context() lack synchronization, allowing Thr
llama.cpp builds b1886 through b7445 contain a race condition use-after-free vulnerability in the LLaMA-Android JNI wrapper where bench_1model() and free_1context() lack synchronization, allowing Thread A to operate on freed memory while Thread B concurrently frees the llama_context. Attackers can exploit this by performing heap spray with attacker-controlled data containing a fake vtable to hijack the vtable pointer at offset +0x30, causing llama_batch_allocr::clear() to dereference arbitrary
Affected packages
- llama.cpp
Exploit signal
- Known exploited (CISA KEV)
- No
- EPSS score
- —
- EPSS percentile
- —
EPSS is the probability a CVE is exploited in the wild; percentile is its rank among all CVEs.
Sources
- https://www.cve.org/CVERecord?id=CVE-2026-70640
- https://github.com/Vladimir-tokarev-cyera/llama-cpp-security-patches
- https://github.com/ggml-org/llama.cpp/commit/5c0d18881e0e9794c96b2602736b758bac9d9388
- https://github.com/ggml-org/llama.cpp/releases/tag/b7446
Structured record: https://ubuntu.com/security/CVE-2026-70640
Recommended response stack
Some links are affiliate links — HackTribune may earn a commission at no extra cost to you.
Get incidents like this as alerts for your stack.
Join the beta