Jumping from a sequence in a text file to nucleotides in a viral capsid is a VERY long road. Anyone with the requisite knowledge, experience, equipment and desire would not be inhibited by a pay-wall or other administrative hoops. They've already jumped though plenty of hoops just to get the equipment.
Also an enterprising mad scientist would likely just splice together genes from viruses already available in the wild. If one wanted to make a "resistant" or "extra-virulent" strain they would probably use random mutagenesis techniques (X-ray and chemical methods). These techniques have already created bacteria and plants resistant to all but the most toxic of poisons. I myself have even made toxin resistant plants using only a standard microwave (and lots of patience).
So no, I would argue that hiding viral sequences would not prevent any mad scientist from creating the next "super-bug".
Well, seed catalogs sell Castor Beans--which can kill. Actually, I saw a really good talk about killer plants recently (totally not bioinformatics http://www.cornell.edu/video/index.cfm?VideoID=918 ). There's plenty of stuff in use that can do that too. Some of it probably in your institution's landscaping.
I don't think that creating a biohazard is easier than creating cures.
Considering how much effort goes into creating solutions and progress is still only one small step at a time it is probably safe to say that (at least for the near future) there is little potential for using the GeneBank info for greater evil.