Working on training a reinforcement learning agent to use Metasploit to interact with computer networks to do autonomous pentesting. I was a pentester for a few years at the same time I was reading Neuromancer, which has made me decidedly obsessed with working on an agent that could do network pentesting. I presented my current results at StrangeLoop[1]. Check it out if you're into infosec or RL.
Planning to clean up the code I've been experimenting with it and release it before the end of the year[2] and show it off at Defcon. Goal is to get more people interested in working on the project. I'll need to be able to generate tons of realistic looking networks with realistically vulnerable machines to capture a large enough distribution for a robust agent.