After reading a research paper titled “Chipping Away at Censorship Firewalls with User-Generated Content” from Sam Burnett, Nick Feamster, and Santosh Vempala of Georgia Tech.
I though of using AI to push the concept further, by creating a huge corpus of innocuous content we can remove the bottleneck implied in the paper. We can create music, images and videos with hidden messages. The protocol is as follow.
The messages are first rateless erasure encoded resulting in many chucks for each message.
We then vector encode the chunks in AI content which is stored anywhere online where user-generated content is accepted.
To coordinate on the location of the chunks, matching pairs of search and publish tasks are created. For example, latest tagged “world cup” tweets, latest “downhill skying” videos on youtube or latest “X” style on “X” AI music storage website. Tasks must be innocuous web searches any user would do, otherwise it’s easy to be discovered.
Messages are mapped to tasks via a consistent hash function so that tasks can be updated smoothly via the covert comm channels themselves.
The resulting channels have interesting properties that other private communication technique don’t. Notably, it can bypass most censorship as long as user-generated content can be hosted.
P.S. For more details, I can share the paper DM me