So, I commented elsewhere on HN: https://news.ycombinator.com/item?id=8819389 https://news.ycombinator.com/item?id=8819389 about the move away from the current

So, I commented elsewhere on HN: https://news.ycombinator.com/item?id=8819389 https://news.ycombinator.com/item?id=8819389 about the move away from the current network security rules we have. I think that network-oriented defenses are not the right semantic for modern applications, as modern applications are no longer tied to physical semantics, or the semantics that modern hardware filters on. ----- I have a different view on security. A lot of people are talking about defense in depth, and security being enforced by the network. I think although that kind of works, it's largely a mistake. Security, as enforced by middleboxes is fundamentally flawed, especially as we see the proliferation of systems that are too complex, and rapidly changing for middleboxes to understand. Although, they can be a tool in laying defensive trenches, they are nothing more than a simple barrier. I think that endpoint security is what needs to be enhanced. Looking at what's been disclosed about malware recently (APTs: https://en.wikipedia.org/wiki/Advanced_persistent_threat https://en.wikipedia.org/wiki/Advanced_persistent_threat), we need to focus on a few things: 1) Isolating components: Components, and processes themselves should be isolated from one another, and the operating system. A web server should be able to run malicious code without fear of it easily getting root access. We have tools to make this possible today, like Apparmor, and SELinux. The fact that distributing Apparmor, or SELinux profiles with applications today isn't normal, makes me sad. This also probably means more unikernels, and lightweight hypervisors. 2) Modularization: We should do our best to split the components of a system up into small, well-understood pieces. Rather than trying to combine multiple components into one monolithic process, breaking up components into their individual components. I think Cloudflare's work around Keyless SSL was great, but we've had HSMs for years -- that could have easily avoided such disasters as Heartbleed. 3) Decent abstractions: With the advent of further distributed systems, and modularization, we're going to see more and more interfaces between systems crop up. Largely, CISOs are limited to what the systems, and networking team can do. These limits typically manifested themselves at layer 4 on the network side of the house, and limited layer 7 capabilities like URL filtering, and file system ACLs. I think we need better stories around resource access and authorization. For years, we've had systems like Kerberos which would have been a great building block, but yet goes unused. I hope we see X509, and GPG become more ubiquitous for access management, and encryption.