I remember when people in the security community started filing CVEs against the TensorFlow project, claiming that code execution was possible with a handcrafted TensorFlow graph, and the team would have to try and explain, "TensorFlow GraphDefs are code".
The whole situation around CVE in Tensorflow is very painful, you get GitHub security notifications for any public repository using TF because of a "known CVE", even though it's basically just a train.py script that is not deployed anywhere.
Comments
I remember when people in the security community started filing CVEs against the TensorFlow project, claiming that code execution was possible with a handcrafted TensorFlow graph, and the team would have to try and explain, "TensorFlow GraphDefs are code".
The whole situation around CVE in Tensorflow is very painful, you get GitHub security notifications for any public repository using TF because of a "known CVE", even though it's basically just a train.py script that is not deployed anywhere.