back to article ChatGPT side-channel attack has easy fix: Token obfuscation

Almost as quickly as a paper came out last week revealing an AI side-channel vulnerability, Cloudflare researchers have figured out how to solve it: just obscure your token size. The paper [PDF], from researchers at the Offensive AI Institute at Israel's Ben Gurion University, found an issue with how all non-Google ChatGPT …

  1. that one in the corner Silver badge

    Specially trained LLMs designed to examine the packets

    Training a Neural Net to examine packets is an approach that can be applied to any data stream that exhibits a bursty nature related to its operation, such as the route through the site that a user has to follow in order to do a specific action - the two-way "conversation" as the site sends new page contents, the user responds...

    Capture a load of interactions with, say, a banking website: you won't be able to read the encrypted data[1] within the stream, but you may be able to figure out that there is a 79% probability this user is trying to get a new credit card. What you then do with that information - send them more adverts about cheap credit from Kickbacks-Be-Us?

    Of course, that 'Net won't necessarily be an LLM, so won't be all sexy and newsworthy.

    [1] if you can them just use that, stop wasting time with training a 'Net!

  2. Groo The Wanderer Silver badge

    I've got an easier fix: avoid that abortion like the plague it is.

    1. This post has been deleted by its author

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Other stories you might like