You are right, when you look at it like that, my comment was a bit of a hijack. The threat model in question assumes that the attackers know that the person in

You are right, when you look at it like that, my comment was a bit of a hijack. The threat model in question assumes that the attackers know that the person in question has a way to auth / is part of an encrypted / limited access system, which is a very sensible assumption (and is included in the police/accused scenario in any case.) I was probably thinking something along the lines of, [this proposed system] + a way for the user to discretely convey to the auth/security system the info that they are being coerced, and the system authenticating them to a bogus user account / set of sensitive data. But this would be cumbersome and very difficult to implement given the design in question, probably. It would have probably been more relevant to mention encryption systems with plausible deniability - e.g. TrueCrypt's hidden volumes [1] and Rubberhose FS [2]. [1]: http://www.truecrypt.org/hiddenvolume http://www.truecrypt.org/hiddenvolume [2]: http://en.wikipedia.org/wiki/Rubberhose_%28file_system%29 http://en.wikipedia.org/wiki/Rubberhose_%28file_system%29