24 April 2008
For a long time I have been sceptical about Private Information Retrieval (PIR) schemes and security schemes based on them. My first experience of PIR was in the single server setting, where communication and computation complexity makes them impractical. Re-reading the The Pynchon Gate I realized that multi-server PIR systems are computationally cheap, bandwidth efficient and relatively simple to implement.
The ‘only’ downside of multi-server PIR is that they are subject to compulsion attacks. A powerful adversary can force servers, after a query, to reveal the client queries, and can infer which document was retrieved. This is an inherent limitation of using a collection of trusted parties, so it is difficult to eliminate. On the other hand a system can make the task of the attacker much more expensive and difficult, though the use of forward security mechanisms.
Here is a proposal for achieving forward-secure compulsion-resistant multi-server PIR: the user contacts the servers one by one, using an encryption channel providing forward secrecy (OTR would work; so would SSL using signed ephemeral DH.) After the result of the query is returned, the server securely deletes all information about the query, and forgets the session keys associated with the channel. At this point an adversary will never be able to retrieve any information about the query or the result, even if they get access to all the secrets on the server.
The user can then proceed to perform the same protocol sequentially with all the other servers participating in the PIR scheme. After sessions with each server close, the user is guaranteed that the query information will never be retrieved in the future. A single honest server, willing to provide strong guarantees against compulsion, is sufficient to guarantee this property, even if all the others log requests and are ready to hand them over to the adversary.
Furthermore the sequential nature of the requests allow a client to terminate the query early, if there is any suspicion that one or more servers act under compulsion. This could be detected through a covert channel, a change of key, or unavailability. This technique is a further argument for operators to terminate their services instead of giving in to compulsion.
21 April 2008
The list of accepted papers for the Privacy Enhancing Technologies Symposium for 2008 (PET) has just been published. The program this year is very heavy on anonymous communication, and traffic analysis, cementing the symposiumas the de-facto venue in these disciplines.
A selection of papers to appear, that make a very interesting read (thanks to the authors giving me a copy!), include:
- Studying Timing Analysis on the Internet with SubRosa
Hatim Daginawala and Matthew Wright (University of Texas at Arlington, USA)
- Perfect Matching Disclosure Attacks
Carmela Troncoso, Benedikt Gierlichs, Bart Preneel, and Ingrid Verbauwhede (COSIC, K.U. Leuven, Belgium)
- On the Impact of Social Network Profiling on Anonymity
Claudia Diaz, Carmela Troncoso (K.U.Leuven, Belgium), and Andrei Serjantov (The Free Haven Project, UK)
- Chattering Laptops
Tuomas Aura (Microsoft Research, UK), Janne Lindqvist (Helsinki University of Technology, Finland), Michael Roe (Microsoft Research, UK), and Anish Mohammed (Royal Holloway, University of London, UK)
- Metrics for Security and Performance in Low-Latency Anonymity Systems
Steven J. Murdoch and Robert N. M. Watson (Computer Laboratory, University of Cambridge, UK)
I should blog in detail about our own contributions, when the final drafts are in, so that I can provide a link to the full text:
- Bridging and Fingerprinting: Epistemic Attacks on Route Selection
George Danezis (Microsoft Research, Cambridge, UK) and Paul Syverson (Naval Research Laboratory, USA)
- How to Bypass Two Anonymity Revocation Schemes
George Danezis (Microsoft Research, Cambridge, UK) and Len Sassaman (K.U. Leuven, Belgium)
Lets not forget that the PET Award for 2008 deadline is still ahead of us, and good papers from the last two years deserve to be nominated for it!