Roughtime (19 Sep 2016)
Security protocols often assume an accurate, local clock (e.g. TLS, Kerberos, DNSSEC and more). It's a widely accepted assumption when designing protocols but, for a lot of people, it just isn't true. We find good evidence that at least 25% of all certificate errors in Chrome are due to a bad local clock.
Even when the local clock is being synchronised, it's very likely to be using unauthenticated NTP. So if your threat model includes man-in-the-middle attackers then you still can't trust the local clock.
There have been efforts to augment NTP with authentication, but they still assume a world where each client trusts one or more time servers absolutely. In order to explore a solution that allows time servers to be effectively audited by clients, myself and my colleague Matt Braithwaite (with assistance and advice from Ben Laurie and Michael Shields) have developed Roughtime.
Very briefly: using some tricks we believe that it's viable to deploy servers that sign a client-chosen nonce and timestamp on demand. Once you have several of these servers, clients can generate their nonces by hashing replies from other servers with some entropy. That proves that a nonce was created after the reply was received. Clients maintain a chain of nonces and replies and, if a server misbehaves, can use replies from several other servers to prove and report it.
Currently there's only one Roughtime service running, so the idea of spreading trust around is inchoate. But we would like to gauge whether others are interested in this idea, specifically whether there are any organisations who would be seriously interested in deploying something like this in their clients. (Because I assume that, if you have clients, then you'll also be interested in running a server.)