That's true. The problem would be if you get attackers who have exploits to break out of virtual machines. What I can say in those regards is you might be able

That's true. The problem would be if you get attackers who have exploits to break out of virtual machines. What I can say in those regards is you might be able to design that out by creating a DSL for specific kinds of exploits and then using that for the first version. You could always have a panel of humans that approve or deny exploits on the network but that kind of thing defeats the point of using the service. What I'm trying to accomplish is to build something without the need for human intervention. Because with a human-run service you open yourself up to politics and bureaucracy and suddenly its the same as with a regular bug program. You've given me something important to think about.