I don’t think the solution to cybersecurity dangers in the age of AI is increasing interconnectedness and decreasing time-to-ship-cve-fixes. Eventually we will HAVE TO start decreasing the attack surface (amount of code in production) and introducing artificial hermetization at a larger scale than we do now (VPNs, IP whitelisting, other infrastructural security measures).
Personally, I already don’t trust the amount of code I have running in my home lab to be secure enough to withstand having the TCP/UDP ports open to the world (or even to a single country). Like most people, I’m using a VPN. A single point of failure, much smaller attack surface.