Mitigating factors include typing style, multi-case passwords, uncommon laptops.

Laptops more susceptible to having keyboard recorded in quieter areas, like coffee shops, libraries, offices. Previous attempts at keylogging VoIP calls achieved 91.7 percent top-5 accuracy over Skype in 2017 and 74.3 percent accuracy in VoIP calls in 2018.

  • Lmaydev@programming.devEnglish
    11·
    1 year ago

    The researchers used a 2021 MacBook Pro to test their concept, a laptop that “features a keyboard identical in switch design to their models from the last two years and potentially those in the future,” typing on 36 keys 25 times each to train their model on the waveforms associated with each key

    • Kinglink@lemmy.worldEnglish
      8·
      1 year ago

      I think my question is more “Does this only work on MacBook Pro” though now I wonder how much the angle the Mac Book is open will matter to the program, or if an external microphone will affect it (or completely throw it off).

      Not trying to tear this down, but more understand the scope needed for this attack to work, or what could break this approach.

      • Lmaydev@programming.devEnglish
        3·
        1 year ago

        They used an iPhone 13 mini, 17 cm away, to record the keyboard’s audio for their first test. For the second test, they recorded the laptop keys over Zoom, using the MacBook’s built-in microphones, with Zoom’s noise suppression set to its lowest level. In both tests, they were able to achieve higher than 93 percent accuracy, with the phone-recorded audio edging closer to 95-96 percent.

        So it can be listened to by a nearby device or over zoom it seems.

    • jpj007@kbin.social
      1·
      1 year ago

      Now I’m wondering if they trained the AI on the exact same machine that they used in their testing. Seems to me that tiny variances in the keyboards of even the exact same model could have significant effects on this sort of thing. And then there’s different levels of wear from usage, crumbs in the keyboard, etc.

      I’d be amazed if it is actually this accurate across all machines of the same model.

      • Lmaydev@programming.devEnglish
        1·
        1 year ago

        Yeah it’s more a proof of concept really. The fact it works at different distances shows it doesn’t have to be identical but there’s a lot of factors to take into consideration.

        Creating a much larger training set with all sorts of variations would be the next step.