Some false-matches on the files can be alleviated, and indeed internally I have that fix ready for the next build. The issue at hand, though, is how much leniency can safely be given. If it helps you have some perspective, the characters tripping it up are patterns such as:
<a blahblahblah >
And I think it’s safe to say that if you have selected that file uploads be sanitized against XSS attacks, that finding what appears to be a link inside one is suspect. The problem is that certain browsers ::cough:: IE6 ::cough:: when they perceive HTML tags within an image, will just serve that image’s “contents” as HTML, ignoring the MIME type that the server sends.
So we’re taking an extremely cautious stance before adding further leniency.