Strict sandboxing andor airgapping render an AI bot nearly useless in practice. If it can't freely make HTTP requests, it is not going to get much done for producing innovative outputs.
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.
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.