Hi Matt,
for security, have a look at RobocodeSecurityPolicy.
There is system property OVERRIDEFILEREADSECURITY. You could take inspiration and implement something similar for networking. I don't know much about the topic, but it seems to me that URLPermission or SocketPermission would be the way to go.
We are open to PR contributions. Note, that such settings would not be enabled on any Rumble server, so such robot would only work in your own environment.
About reinforcement learning, please share your journey with us. I'm definitely very interested.
Cheers
Pavel