CVE-2021-37686: 
Python vulnerability analysis and mitigation

TensorFlow is an end-to-end open source platform for machine learning. In version 2.6.0, the strided slice implementation in TFLite contained a logic bug which could allow an attacker to trigger an infinite loop. This vulnerability was assigned CVE-2021-37686 and was discovered in August 2021 (GitHub Advisory).

The vulnerability arose from newly introduced support for ellipsis in axis definition in the TFLite strided slice implementation. The bug occurred in a loop where an attacker could craft a model such that ellipsis_end_idx would be smaller than i (e.g., always negative). In this case, the inner loop would not increase i and the continue statement would cause execution to skip over the preincrement at the end of the outer loop, resulting in an infinite loop (GitHub Advisory).

The vulnerability could allow attackers to trigger an infinite loop in TensorFlow Lite, potentially causing a denial of service condition in applications using the affected version (GitHub Advisory).

The issue was patched in TensorFlow version 2.6.0 through GitHub commit dfa22b348b70bb89d6d6ec0ff53973bacb4f4695. Users should upgrade to the patched version. TensorFlow 2.6.0 was the only affected version (GitHub Advisory).

The vulnerability was reported by members of the Aivul Team from Qihoo 360, demonstrating ongoing security research in the machine learning framework space (GitHub Advisory).