All such measures make the bot useless. If it can't make HTTP requests or call APIs, it is useless.

What is perhaps rationally needed is the an integrated AI based security layer that observes the stream of requests and occasionally firewalls them if an explanation is not provided. This parallel layer must be able to bidirectionally communicate via chat with the primary agent.