Signal president Meredith Whittaker said AI agents embedded within operating systems are eroding the practical security guarantees of E2EE.
      • scytale@piefed.zipEnglish
        10·
        15 hours ago

        I randomly met one of their devs at a basketball pickup game. At the time he was a new hire and he said his primary role would be to work on the desktop app. He said they were only 2 people in the team as of his joining. I told him the app in its current state (at the time) was still so basic and definitely needs more attention.

    • deliriousdreams@fedia.io
      9·
      15 hours ago

      Android has a market share of something like 42.5%. Google is adding Gemini to devices I’m in a way that gives Gemini core android apps permissions. Right now these settings can be changed, but the default settings give this access.

      It’s not just MS.

        • deliriousdreams@fedia.io
          1·
          6 hours ago

          Gmail: Click the gear icon in the top right of Gmail > See All Settings. From there, turn off Smart Compose, Smart Compose Personalization, and Smart Reply. (There’s also the Smart Features setting, which turns off everything even remotely AI-related, but that will also disable spelling and grammar check.)

          On device: Open the Gemini app on your Android. Tap your profile icon in the top-right corner. Go to Gemini Apps Activity. Tap Turn off Turn off and delete activity, and follow the prompts. Select your profile icon again and go to Apps. Tap the toggle switch to prevent Gemini from interacting with Google apps and third-party services. Avoid using Gemini Deep Research with Gmail, Drive, or Chat (these sources are turned off by default). If you have already selected them when enabling Deep Research, open Sources* and clear their checkboxes.

          Once that’s done, disable the app or remove it if your device allows that.

      • Butterphinger@lemmy.zip
        2·
        7 hours ago

        Anonymous methods are known. You can retain anonymity regardless of what Fed can peek into the box. We accept this as a necessary evil to XMPP rooms.

        1/3 of the room might be running Ai agents on their screens, so, the international spy network knows what everyone’s doing… but they have no fucking clue who the other 2/3 is, nor is the history constantly browsable, mostly pieces at a time.

        In a room with friends who know who you are, the relationship would become more obvious.

      • sp3ctre@feddit.org
        11·
        20 hours ago

        Sooner or later, I see myself forcing my important contacts to use linux. But indirectly, like, if they want to keep messaging me, they need to switch to a trustworthy OS. Of course I’ll have to verify beforehand if they actually use it.

    • lad@programming.devEnglish
      8·
      19 hours ago

      In O’Reilly’s Clawdbot research, he identified hundreds of exposed control panels reachable over the public internet, some lacking any authentication. These interfaces provided access to full conversation histories, API keys, OAuth tokens, and command execution features across services including Slack, Telegram, Discord, WhatsApp, and Signal. In several instances, Signal device-pairing data was stored in plaintext, enabling attackers to take over accounts remotely.

      Sounds like people can set it up on their own in any OS, but I admit that I didn’t exactly understand what that control panels are

      • scytale@piefed.zipEnglish
        3·
        15 hours ago

        Yeah, in this case the OS doesn’t really matter if the person willingly (or unknowingly) installs an AI agent (assuming it’s compatible with linux as well).

      • JubilantJaguar@lemmy.world
        6·
        18 hours ago

        Was just about to quote the same passage with the same question! Seems we’re not familiar with what the kids are doing with AI.

        Before, normies didn’t know what they were doing and so they couldn’t do it. Now they can go ahead and do it anyway. Technical ignorance is about to become a serious problem.