Theo and the OpenBSD people have never been big fans of virtualization or mandatory access control systems being used for security. This is their opinion and it isn't shared by all expert level security researchers, it is actually fairly controversial in security circles that OpenBSD has no MAC system and I personally find it annoying that they have very little virtualization support. Qubes is an OS that was also made by expert level security researchers and they focused on the isolation route. Really that is all that you need to do? To hack the system all you need to do is compromise the system? That list of steps adds no additional information and is inherently obvious. The thing is to break out of the jail to virtualbox, or to break out of virtualbox to the host system, the attacker needs to find an additional vulnerability. Finding an additional vulnerability in the hypervisor of virtualbox or vmware may not be an extremely difficult thing to do because they use overly complex code bases. The Qubes people went with Xen and I think it probably has a less complex and more correct hypervisor that will be harder to break out of. SEL4 is a formally verified microkernel, some security experts have said that microkernels are essentially hypervisors, so if the assumptions the SEL4 proof is based on hold to be true and everything was correctly formally verified, it is (I think) provably impossible to break out of the isolation provided by SEL4. That goes back to what I said before about having a correct hypervisor. The only real argument against using virtualization in this way is that an attacker can break out of the isolation if they find an additional vulnerability, but guess what they still need to find/spend an additional zero day which buys you time in which to detect them and if you use a correct hypervisor they wont be able to break out of it anyway. VMs are coded with security in mind as one of the potential uses. Sandboxes are a wide known security technique. Java uses a VM that gives it security advantages over non-interpreted languages. VMs and security go hand in hand actually. VM are also used for analysis of viruses and they are used so the virus can not effect the host system. VMs are also used in hosting environments so if one customer is pwnt the attacker can not as easily damage other customers. OpenBSD is not guaranteed security, it hasn't even been formally verified so there is not a proof of correctness. It is highly correct though because it has been audited in depth by a large number of security experts and really good coders. OpenBSD probably has less vulnerabilities left in it than anything else that hasn't been formally verified. Of course when you use OpenBSD you will be installing all kinds of additional things that have not been nearly as highly audited and could still contain remote code execution vulnerabilities, such as firefox. OpenBSD gives an additional layer of protection from these applications being attacked in the form of ASLR and it has two tools for isoation Systrace and chroot, but it doesn't have a mandatory access control system and it doesn't support much virtualization technology. This is why I prefer to run OpenBSD as a virtual machine guest instead of the host system. Well you can virtualize a 64 bit OS if you have a CPU that supports hardware virtualization. Of course OpenBSD isn't as secure on 32 bit systems, ASLR on such such systems can be brute forced. The same exact problem is created by using a 32 bit OS in any case, ASLR on 32 bit OS is not secure. OpenBSD has a more secure version of chroot than most other operating systems. And yes if you are up against the CIA or NSA they will cut through all the layers of isolation that you throw at them probably near instantly and without being detected, but if you are up against script kiddies to intermediate hackers a single layer of isolation with virtual machines will imo probably be enough to prevent them from breaking out of isolation. If you are up against advanced hackers who are not CIA or NSA level they will still probably eventually be able to break out of all layers of isolation that you throw at them, but in this case isolation still buys time in which you can detect them with intrusion detection software. Please provide me with logs from Theo or any citation that claims it is not safe to use encryption in a virtual machine. Also I have never heard of virtualization decreasing security before, although I have heard from a number of security experts that vulnerabilities in hypervisor and other things can allow an attacker to break out of the isolation if they can discover and exploit the vulnerabilities. I have also heard from security experts that isolation is the best current technique for security and correctness and randomization are not currently at a state where they can be relied on. I have also heard from security experts that isolation like this is worthless and correctness is the only way to go (of course I guess we just need a correct hypervisor for virtualization based isolation like this to be the way to go then). Yes I agree that hardware based isolation is better but ignoring virtualization based isolation is a bad mistake imo. Banks use hardware chips with isolation layers and they have still been penetrated by hackers. But just because the most leet hackers in the world can pwn banks doesn't mean they should throw in the towel and use no security technology at all. Yes there have been OpenBSD zero days before and no there is no mathematic proof that there never will be again. Also you run unaudited applications in OpenBSD all the time. Also if you add layers of isolation you buy more time in which to detect an intrusion, intrusion detection software and isolation go hand in hand. Show me any citation about virtual machines and encryption leaks. Also there could be a vulnerability in the code of the pf firewall, so it shouldn't be used because a leet hacker might be able to break out of the isolation is provides .