I am wary of software security advice that leads with "don't trust user input", or revolves around "validate user input". That principle has been the core of software security strategy for going on 20 years, and has bought us very little. In the real world, we have to start by acknowledging that the verb "trust" is situational, and that in some circumstances virtually all user input is "trusted" somehow.
You could phrase this less tactfully as "Validate user input? No shit? Now what?"
Here's some software security advice I'd like to offer, as a software security practitioner to the startup stars and the Dr. Drew Pinsky of HN (ducks):
* Plan to update your platform software at inconvenient times. Dry-run your update process, so you know it will work on no notice. I personally believe you should also avoid your OS's packaged versions of things like Apache and nginx; having a known-working source build gives you control of when and how you'll apply patches; it's something you should be able to do easily.
* Put someone on your team in charge of tracking your dependencies (C libraries, Ruby gems, Python easy_install thingies) and have a process by which you periodically check to make sure you're capturing upstream security fixes. You should run your service aware of the fact that major vulnerabilities in third-party library code are often fixed without fanfare or advisories; when maintainers don't know exactly who's affected how, the whole announcement might happen in a git commit.
* Use TLS to encrypt data in motion and use GPG to encrypt data at rest, and don't do any other kind of crypto. GPG blobs are large and expensive looking, but getting custom crypto right is also very expensive.
* Stay on your platform's "golden path" for web security issues. Rails developers should default-whitelist ActiveRecord models, enable CSRF protection, and avoid "html_safe-ing" strings so that the maximum amount of code inherits default protections against mass assignment, XSS, and CSRF. Anything you customize will probably bite you in the ass. Keep your code boring.
* Triple check any piece of code that "shells out" to command line tools. By "triple check": have a process by which three signoffs are required to merge any such code into the deployment branch.
* Be extraordinarily wary of library code for web apps that includes "native" C/C++ code. Very popular C library code for modern web platforms has been found susceptible to basic memory corruption issues, because the kinds of people that look for memory corruption bugs don't usually think to troll Github for Ruby, Python, and PHP code with native backend code; terrible bugs can thus stay latent for years in code you can point a URL to and read.
* People will hate me for saying this but I'm here to offer honest advice: prefer almost any modern language to PHP or Perl. I don't know what to tell you other than that PHP and Perl apps fare worse on security assessments than everything else.
* I have more than once recommended that people who are very very concerned about platform security (ie, about the likelihood that there are memory corruption bugs in their language stack) use JVM languages.
* Do your admin stuff out-of-band. Write a separate admin app (bonus: the admin app can look shitty, and so is less expensive to maintain) that requires a VPN connection to access. Avoid special-privilege accounts in your normal apps. From years and years of experience working with startups: this is something you will mess up on.
* Triple-check code that handles direct file uploads and downloads. The filesystem introduces a new namespace, so upload/download code needs to juggle different privilege and authorization domains to handle it. We see fewer problems at companies that dump blobs with opaque names into S3 than we do with apps that have a file repository with named files.
* HAVE A SECURITY PAGE FOR YOUR APP. Have that page very cordially invite people to submit security flaws to an email address at your site; provide a PGP key for it. If I was maintaining a commercial app, I'd put a phone number on that page too. If you don't have this page, you should know that you are tacitly inviting people to report security flaws to Twitter.