only criminals will have privacy. I guess we can't emphasize more. When you think your cellphone conversations are encrypted, the termination is at service provider's gateway. The SP has access to the clear voice. Reading e-mail over SSL? Again the email server is storing the plain text of the email. And people who are out there to get you will arm-twist the service providers to provide access. Recent example is governments demanding keys to blackberry communication. So what do we need for complete privacy? Well, end to end encryption. Right from lips to ear. If the message is decrypted anywhere else in the channel, that's the point agencies can tap into. It should be like users conversing in Greek/Latin/Hebrew in a public place, everyone can hear them but have no clue what's being talked about.
An implementation of PGP in the voice world might have been a good choice. At least in principle. What we need is a voice sample encrypting machine, that sits right in between the DSP and the phone's voice processing interface. Actually I'd prefer a headphone/mic combination where they already encrypt voice. Two difficulties in that, though. It has to setup encryption key for a session, for which it has to authenticate and negotiate with a co-operating peer, and it'll have 6ms to cryptographically and signal-wise process each voice packet (assuming 40kHz sampling, stereo input and output). But such chips were available even 5 years ago.
Though all the bells and whistles of PGP aren't needed all along. The problem to solve first is to setup a session key over untrusted channel. Diffie-Hellman being a sitting duck to man-in-the-middle, you need something out of band negotiated with the recipient (or their certificates). Obviously a long term shared key might threaten your perfect forward secrecy, where if the key is broken at any time, all your previous conversations gets compromised. That way, the line taken by OTR is much acceptable, they also purposefully avoid non-repudiation. So something like that should be available for voice streams. The conversation will stay confidential, and even when one party is held at gunpoint, it is not possible to conclusively prove the identity of the peer.
But they may say using encryption is incriminating enough.
An implementation of PGP in the voice world might have been a good choice. At least in principle. What we need is a voice sample encrypting machine, that sits right in between the DSP and the phone's voice processing interface. Actually I'd prefer a headphone/mic combination where they already encrypt voice. Two difficulties in that, though. It has to setup encryption key for a session, for which it has to authenticate and negotiate with a co-operating peer, and it'll have 6ms to cryptographically and signal-wise process each voice packet (assuming 40kHz sampling, stereo input and output). But such chips were available even 5 years ago.
Though all the bells and whistles of PGP aren't needed all along. The problem to solve first is to setup a session key over untrusted channel. Diffie-Hellman being a sitting duck to man-in-the-middle, you need something out of band negotiated with the recipient (or their certificates). Obviously a long term shared key might threaten your perfect forward secrecy, where if the key is broken at any time, all your previous conversations gets compromised. That way, the line taken by OTR is much acceptable, they also purposefully avoid non-repudiation. So something like that should be available for voice streams. The conversation will stay confidential, and even when one party is held at gunpoint, it is not possible to conclusively prove the identity of the peer.
But they may say using encryption is incriminating enough.