CVE-2021-37684: 
Python vulnerability analysis and mitigation

TensorFlow, an end-to-end open source platform for machine learning, was found to contain a vulnerability in its TFLite pooling operations. The vulnerability (CVE-2021-37684) was discovered in versions prior to 2.6.0, where the implementations of pooling in TFLite were susceptible to division by zero errors due to missing checks for divisors being non-zero (GitHub Advisory).

The vulnerability stems from the lack of validation checks in TFLite pooling operations, where the implementation did not verify that divisors were non-zero before performing calculations. This oversight could lead to division by zero errors in the affected versions. The vulnerability was assigned a Low severity rating (GitHub Advisory).

When exploited, this vulnerability could cause a denial of service condition through division by zero errors in applications using TFLite pooling operations (GitHub Advisory).

The issue was patched in GitHub commit dfa22b348b70bb89d6d6ec0ff53973bacb4f4695 and included in TensorFlow 2.6.0. The fix was also backported to versions 2.5.1, 2.4.3, and 2.3.4. Users are advised to upgrade to these patched versions to mitigate the vulnerability (GitHub Advisory).