You might find it amusing to google perfect forward secrecy. With PFS, they can't decrypt previous intercepts even if Google were to disgorge the private key in effect at the time.
But the truth is, you can't trust any channel unless you and your correspondent do your own encryption. Any encryption done by the channel is subject to compromise.
One-time-pads provide secure encryption, PROVIDED that the numbers are genuinely random, and that they are used ONLY ONCE. The problem with them is distributing them to people to whom you want to send encrypted messages.
"Random numbers" generated by a computer are really pseudo-random. They can satisfy tests of randomness (each digit appears approximately one-tenth of the time, each pair one-hundredth of the time, autocorrelation is very small, etc.) However, they eventually repeat. Genuinely random numbers never repeat. The repetition makes it comparatively easy to break the encryption of long messages.
I've tested "random" numbers obtained from tables of statistical data, such as population of cities, from sources such as the STATISTICAL ABSTRACT OF THE UNITED STATES. They not only pass randomness tests but have low autocorrelation, and don't repeat. You need to agree on page number, line number, etc. so the recipient knows where to start picking numbers.
Another alternative is to get a set of "gamers dice," i.e. 10-sided dice, and generate your own random numbers. Again, you have to distribute the one-time-pads to the people with whom you want to communicate.
Getting a big "supply" of random numbers is not easy, but it can be done with some degree of effort. Do it before you need it.