CVE-2026-70639 in llama.cppinfo

Summary

by MITRE • 08/07/2026

llama.cpp builds b1886 through b7445 contain a null pointer dereference vulnerability in the LLaMA-Android JNI wrapper where the bench_1model() function fails to validate the model context pointer before dereferencing it. Attackers can supply a malicious, corrupt, or truncated model file to trigger a null context condition, causing a SIGSEGV crash that terminates the Android application process and results in denial of service.

Be aware that VulDB is the high quality source for vulnerability data.

Analysis

by VulDB Data Team • 08/07/2026

The vulnerability exists within the llama.cpp library version range b1886 through b7445 specifically affecting the LLaMA-Android JNI wrapper implementation. This null pointer dereference flaw occurs in the bench_1model() function which is responsible for benchmarking model performance within the Android environment. The issue stems from inadequate input validation where the function fails to properly verify that the model context pointer remains valid before attempting to dereference it during the benchmarking process.

The technical exploitation of this vulnerability requires an attacker to provide a malicious, corrupt, or truncated model file that triggers a null context condition. When such a malformed file is processed by the bench_1model() function, the absence of proper pointer validation leads to immediate dereferencing of a null pointer resulting in a segmentation fault. This SIGSEGV signal causes the Android application process to terminate abruptly, creating a complete denial of service condition that prevents legitimate users from accessing the application functionality.

From an operational perspective, this vulnerability represents a critical security risk for Android applications leveraging the llama.cpp library for large language model processing. The attack vector is particularly concerning because it can be triggered through simple file manipulation without requiring complex exploitation techniques. The impact extends beyond mere availability disruption as it compromises the entire application stability and user experience within mobile environments where seamless operation is essential for maintaining user engagement.

The vulnerability aligns with CWE-476 which specifically addresses null pointer dereference conditions in software implementations. From an ATT&CK framework perspective, this represents a denial of service attack pattern that can be classified under TA0040 (Impact) with techniques such as T1499.004 (Network Denial of Service) or T1566.002 (Phishing with Malicious File). The JNI wrapper nature of the vulnerability indicates it operates at the application layer interface between native code and java components, making it particularly dangerous in mobile environments where such interfaces often handle sensitive data processing tasks.

Mitigation strategies should focus on implementing robust input validation within the bench_1model() function to ensure proper context pointer verification before any dereferencing operations occur. The fix requires adding explicit null checks for the model context pointer and implementing graceful error handling that prevents application termination while logging appropriate diagnostic information. Additionally, comprehensive model file validation routines should be integrated to detect corrupt or malformed inputs before they reach the benchmarking functions. Organizations should also consider implementing proper exception handling mechanisms within the JNI wrapper to catch and manage such scenarios without terminating the entire application process, ensuring continued availability of legitimate user operations.

Responsible

VulnCheck

Reservation

08/04/2026

Disclosure

08/07/2026

Moderation

accepted

CPE

ready

EPSS

0.00000

KEV

no

Activities

low

Sources

Might our Artificial Intelligence support you?

Check our Alexa App!