That multisignature/blockchain commitment idea seems really solid, Peter.

Thanks very much indeed everyone, this is all very helpful. Much to research and think about.

Interestingly, a thread is presently raging on liberationtech about Tor Browser Bundle, and the subject of automatic updates has come up. Gregory Maxwell responded thusly (cross-posting for completeness):

> _please_ don't deploy automatic updates in a sensitive environment
> like this without at least quorum signatures (like gitian downloader)
> and timed quarantine with negative signatures (harder to make strong
> absent a jamming proof network).

-wendell

grabhive.com | twitter.com/grabhive | gpg: 6C0C9411

On Aug 5, 2013, at 7:49 PM, Peter Todd wrote:

> Gregory Maxwell had some good ideas along these lines at the san jose conference. Extending gitian with these kinds of features would be a good approach.
> 
> But I think its worth thinking about attack models. A huge danger with auto-updating is that it is easy to target individuals; if I leave auto-updates on I am essentially trusting the developers capable of signing an update not to specifically try to attack me in the future, a much more risky thing to do than simply  trusting them not to release a malicious release.
> 
> Sure you can try to implement anonymous downloads and similar mechanisms, but they all tend to be fragile with regard to deanonymization attacks.
> 
> A better way is to ensure that the act of making a release available for download must be public, even if you can control what binaries are made available to a particular target. You can do this by putting a commitment in the blockchain itself. Each person on the signing list creates a transaction with a special form from a specific pubkey that commits to the digest of the binaries, and the auto-update code refuses to update unless it sees that special transaction with a sufficient number of confirmations. The developers now can't make a special release for a specific target without letting the world know they did so, even under coercion.
> 
> They developers could of course still make a release with code inside targeting a specific individual, but in theory at least the public can check if their builds are reproducible, and start asking questions why not?