The vulnerability in llama.cpp (builds b1886 through 0.9.0) involves an integer overflow in the LLaMA-Android JNI wrapper's new_1batch() function. Specifically, the function multiplies sizeof(llama_seq_id) by the n_seq_max parameter, which is attacker-controlled, without checking for overflow. This results in heap buffer allocation wrapping around and allocating less memory than required. An attacker can exploit this by supplying a crafted n_seq_max value via a malicious model file or JNI call, causing heap corruption that may lead to denial of service or arbitrary code execution on affected Android applications.

Successful exploitation can cause heap corruption due to insufficient memory allocation, leading to denial of service or arbitrary code execution on Android applications using the vulnerable LLaMA-Android JNI wrapper. The CVSS 4.0 score is 8.5 (high severity), reflecting the potential for significant impact on confidentiality, integrity, and availability.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. No official fix or patch links are currently available. Until a patch is released, avoid using affected versions or restrict usage of untrusted model files or JNI calls that could supply malicious n_seq_max values.