Security and privacy of this service

InkblotPassword.com is a research project deployed by Microsoft Research. It is for demonstration and research purposes only. You are welcome to try it out, but we make absolutely no promise that our implementation will protect your password. Don't use your account here to protect any data you care about, from money to your reputation. We also make no promise that the site will continue running. Should the service prove successful, Microsoft may consider offering the service as a commercial product or service. For now, consider it an unreliable, insecure service run by a couple research coneheads in their spare time, and trust it accordingly.

In both experiments, users missed at most one association, even after having not used the system for one week. Thus it may be advisable to modify the system to allow for successful authentications when k out of a possible n associations are correct. Assuming that all blots produce an equal distribution on responses, this reduces the security of passwords to the level of the original system with only k blots. Therefore, it might be advantageous for users to have to enter associations for more blots. A disadvantage of this approach, however, is that authentication would take longer.

Our preliminary data suggest that inkblot authentication offers a potentially significant improvement over existing widely-deployed user authentication mechanisms. In addition to gathering our quantitative results, we also asked users who had taken part in our experiments for their comments on the system. In almost all cases we received the same response: the users were happily shocked that they could remember such a "huge password." In fact, many users asked if there were any plans to allow the use of the system in their production environment. This kind of positive user experience is arguably as important to the eventual adoption, acceptance and scrupulous use of an alternative password system as any measure of security. More experiments would help confirm or discount our security and memorability results, and could answer such questions as: How many inkblots (that is, how much entropy) can be used before the resulting passwords are no longer memorable? What is the best way to help users retain their inkblot associations? What inkblot-to-character hash function generates the most entropy without sacrificing ease of use? And what inkblot generation algorithms create inkblots with the highest-entropy (or the fewest low-entropy) association spaces?
While inkblot authentication should be quite easy to deploy in a wide variety of settings, there exist some environments (such as devices with tiny screens) where it is unworkable, and alternatives are needed. Adapting the inkblot password scheme to other password-using contexts, such as those in which the user interface is under the control of a (possibly uncooperative or legacy) application, may also require some innovative thinking.