NULL pointer dereference in Ggml Llama.cpp
CVE-2026-70640
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 memory and achieve remote code execution.
Vulnerability class: Race Condition
Published · last modified .
CVSS v3 metric
CVSS v3 base score 7.0 (High). Vector: CVSS:3.1/AV:L/AC:H/PR:N/UI:R/S:U/C:H/I:H/A:H.
EPSS exploit prediction
EPSS: 0.002 (15.2th percentile), scored .
Very low probability of exploitation in the next 30 days; routine patching cadence is appropriate. 15th percentile — 15.2% of CVEs in the catalogue have a lower EPSS than this one. How to read EPSS.
EPSS over last 30 days · oldest: 0.002 · newest: 0.002 · change: 0.000
Affected products
- Ggml Llama.cpp
- Ggml-Org Llama.cpp — versions b1886, 0.9.0, 5c0d18881e0e9794c96b2602736b758bac9d9388
Weakness classification (CWE)
References
- github.com/Vladimir-tokarev-cyera/llama-cpp-security-patches (related, Patch, Third Party Advisory)
- github.com/ggml-org/llama.cpp/commit/5c0d18881e0e9794c96b2602736b758bac9d9388 (Patch, patch)
- github.com/ggml-org/llama.cpp/releases/tag/b7446 (release-notes, Patch, Release Notes, Vendor Advisory)
- www.vulncheck.com/advisories/llama-cpp-b1886-b7445-race-condition-use-after-fre… (Exploit, Third Party Advisory, third-party-advisory)
Frequently asked questions
- What is CVE-2026-70640?
- CVE-2026-70640 is a high-severity vulnerability in Ggml Llama.cpp, classified under Concurrent Execution using Shared Resource with Improper Synchronization (Race Condition). CVSS score: 7.0/10. Published 2026-08-06.
- How severe is CVE-2026-70640?
- High severity. CVSS v3 base score is 7.0 out of 10.