llama.cpp versions b1886 and 0.9.0 contain a race condition use-after-free vulnerability in the LLaMA-Android JNI wrapper. The lack of synchronization between bench_1model() and free_1context() allows Thread A to operate on memory freed by Thread B concurrently. Attackers can exploit this by performing a heap spray with attacker-controlled data containing a fake vtable to hijack the vtable pointer at offset +0x30. This causes llama_batch_allocr::clear() to dereference arbitrary memory, enabling remote code execution.

Successful exploitation allows an attacker to achieve remote code execution by hijacking the vtable pointer and causing arbitrary memory dereference. This can compromise the affected system running vulnerable versions of llama.cpp, potentially leading to full system compromise or unauthorized code execution within the context of the application.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, avoid running vulnerable versions b1886 and 0.9.0 of llama.cpp in environments where untrusted input can trigger the JNI wrapper. Monitor vendor channels for updates and apply patches promptly once released.