'Signal' President and VP warn agentic AI is insecure, unreliable, and a surveillance nightmare

Comments

• wesker@lemmy.sdf.org ⁨4⁩ ⁨days⁩ ago

  Meredith Whittaker is bae.

  source

  • iltoroargento@lemmy.sdf.org ⁨3⁩ ⁨days⁩ ago

    You’re definitely not the only one with that crush lol

    source

    • A_norny_mousse@feddit.org ⁨3⁩ ⁨days⁩ ago

      👋

      source
  • pfr@piefed.social ⁨3⁩ ⁨days⁩ ago

    You’re not wrong. But save some love for Udbhav Tiwari

    source
  • HonoraryMancunian@lemmy.world ⁨3⁩ ⁨days⁩ ago

    Sigourney Weaver vibes

    source
• NuXCOM_90Percent@lemmy.zip ⁨3⁩ ⁨days⁩ ago

  Yes and no.

  At its very core? “Agentic AI” is about the idea of having a bunch of different “agents” communicate with one another in a network with defined(-ish) communication pathways. This is an “agent network”. And if that sounds like microservices/task graphs/how every fucking app works then… you win the No-Prize!!

  And, in that regard, it isn’t any difficult. This service has access to that database. It always has. Hell, this service might still have zero “AI” in it but count as an “agent” for marketing purposes. If the credentials are checked and passed in an appropriate and authorized way, it is as safe as it ever has been. Which… is a different depressing discussion.

  The issue comes into play when you are looking at people rapidly rewriting existing infrastructure just to say they did. And doing so with generative AI that they fundamentally can’t vet (even if they wanted to). THAT is how you break things and THAT is how you introduce new CVEs.

  The issue isn’t that you have this data stored in a SQL table that is accessed by that service which was pre-seeded with credentials in a secure way. The issue is that you have no rewritten both that service and the SQL server in a way that “optimized” things by removing that costly security check.

  source

  • pinball_wizard@lemmy.zip ⁨3⁩ ⁨days⁩ ago

    Hell, this service might still have zero “AI” in it but count as an “agent” for marketing purposes.

    True. I do think that’s where we might find something useful after the bubble pops.

    But it’s some shit marketing though.

    “Microservices, but sometimes it hallucinates.” is a sales pitch targeted for gullible suckers.

    source

    • olympicyes@lemmy.world ⁨3⁩ ⁨days⁩ ago

      “Microservices with delegated responsibility and elevated permissions, but sometimes hallucinates” is slightly more accurate. Claude Cowork probably won’t rm -rf / but what do I know.

      source
  • Slotos@feddit.nl ⁨3⁩ ⁨days⁩ ago

    The windows have all the modern anti-burglary features and the plumbing is immaculate

    — excerpt from a sale advert for a house with its front wall missing and the living room exposed to the elements

    source
  • chirospasm@lemmy.ml ⁨3⁩ ⁨days⁩ ago

    Don’t know why you’re getting downvoted, but you’ve got a solid pulse on the issue, here.

    source
• GhostPain@lemmy.world ⁨3⁩ ⁨days⁩ ago

  China just over there letting them mfers fuck all their own shit up.

  source

  • huquad@lemmy.ml ⁨3⁩ ⁨days⁩ ago

    You never let your enemy know when they’re making a mistake, or several dozen

    source
• pinball_wizard@lemmy.zip ⁨3⁩ ⁨days⁩ ago

  I’ve wiped better ideas off of the used shoes of an old bash profile than the average agentic AI solution.

  source

  • Septimaeus@infosec.pub ⁨3⁩ ⁨days⁩ ago

    *visibly recoils*

    source
• noxypaws@pawb.social ⁨3⁩ ⁨days⁩ ago

  The word “agentic” makes me want to rip my tongue out of my mouth.

  Not only is it a stupid fucking word but it’s unpleasant to even think about pronouncing.

  source
• Sanctus@anarchist.nexus ⁨3⁩ ⁨days⁩ ago

  Yeah? And why should some wagie care that his corp gets hacked through the agent? Or that Microft is watching? This is true but its not an angle to get at anything with because the country where this is being mass produced has an apathetic population

  source

  • evol@lemmy.today ⁨3⁩ ⁨days⁩ ago

    Upper management in my company constantly asks for instances of AI being used. People who align with this whole AI mess get much easier promotion timelines and visibility. My team has decided to realign ourselves to do more ML work partially for this reason. Its even dumber when you realize all these tech execs are themselves invested in these AI companies

    source
• adespoton@lemmy.ca ⁨3⁩ ⁨days⁩ ago

  Meanwhile, they just started publicizing their own e2e encrypted chatbot.

  source

  • crank0271@lemmy.world ⁨3⁩ ⁨days⁩ ago

    The Signal Foundation did? Or Moxie Marlonspike, who I don’t think is technically involved with them anymore? (Not calling you out, I’m not particularly sure of his status there anymore.)

    source
  • Routhinator@startrek.website ⁨3⁩ ⁨days⁩ ago

    You’re mixing the Signal Foundation up with the original creator of Signal who has stepped down and is no longer involved… And is off…

    checks notes

    Building fucking AI bots apparently? Thank goodness he’s not involved with Signal anymore.

    source
• Cocodapuf@lemmy.world ⁨3⁩ ⁨days⁩ ago

  I’m surprised to learn that the CEO is not Moxie Marlinspike, my understanding is that he created the signal protocol. Perhaps he has nothing to do with the apps that implement signal?

  source

  • WhyJiffie@sh.itjust.works ⁨3⁩ ⁨days⁩ ago

    he stepped down a while ago

    source
  • moonshadow@slrpnk.net ⁨3⁩ ⁨days⁩ ago

    He did not singlehandedly create the “signal protocol” (double ratchet key exchange tyvm), that’s not really how it works. He was an early contributor who became an advocate and figurehead because he was a good communicator/unafraid of the exposure. Check out some of his pre-signal sailboating stuff for a good time :)

    source