Meredith WhittakerYes, I did sound the alarm on agentic AI's privacy threat, and rightly so.
https://observer.com/2025/07/signal-meredith-whittaker-agentic-ai-risk/
bengo@Mer__edith thank you 🙏🏼
Viss@Mer__edith i totally agree with you here. i think that at present, "agentic ai" is going to basically eat itself. we constantly see articles about how great it is and how well it does in various scenarios, but what they are omitting in all of those writings is just how fucking insanely bad it is the 9000 other attempts they made before they got "the one that was screenshot worthy".
its the same sleight of hand jobs did with the 1st iphone.
and its real bad.
Scott Francis@Viss @Mer__edith The big question is whether it will end up evolving like iPhone did into something world-changing, and if it will burn down the world before we get to that point
oscarfalconThe thing is How in the hell do we as users protect ourselves from all the"ai" that is being pushed through every damn application? it's why I carry a 4g dumbphone and only do old school in-person important stuff, I don't even use the computer's or phone's calendars or notes applications, none of it feels safe anymore.
Caden@oscarfalcon @Mer__edith not sure what your threat model is, but if you think you'd benefit from a "smart" phone with better privacy, running GrapheneOS and some reliable open source apps like the Fossify apps isn't a bad option. With GrapheneOS you can revoke network access from them at the OS level.
Okuna@Mer__edith the problem, as it appears to me is, that even if I do not use and want it, the other side may still do so and hence use my chats, my pics, my data with the AI at hand, sharing it with Meta, Google, Apple or whoever… How can I protect myself from this? I can’t. I only see two ways:
I speak with each one of my contacts and ask them to respect my privacy. It took me years to get everyone on Threema, now I have to do this again
Second: the providers of the messenger apps should have an indicator on my chat if my opponent uses AI. Not really and not always possible and I am sure Meta will never do this.
I speak with each one of my contacts and ask them to respect my privacy. It took me years to get everyone on Threema, now I have to do this again
Second: the providers of the messenger apps should have an indicator on my chat if my opponent uses AI. Not really and not always possible and I am sure Meta will never do this.
@Okuna this is why we're calling for developer-level opt outs, among other things. Not just per-user consent/opt-in.
@Mer__edith what does developer opt out mean? Sorry, I am not a techie. BTW: Thanks for all the great work your and your team are doing.
@Okuna No worries! This means that we, at Signal--the ones developing the app--should be able to opt-out as Signal. To say, "no data access for you, Agentic AI." Instead of leaving that to each person who uses Signal (which invites the issues you rightly name--the fact that one person ultimately makes that choice for more than themselves, in a way that could still compromise their interlocutors' privacy).
@Mer__edith that sounds great. How could you prohibit, outside agentic AI, prohibit users copying and pasting the entire chat to an LLM for summarization or others?
junkman@Okuna
You're asking for perfection.
What's stopping someone from recording a phone screen with another phone?...
You, the user. If that is bypassed, then it's a whole different problem.
@junkman @Mer__edith I guess so, but again, I am not a techie and have no idea what is possible and what not
rosbeefandino
veroandiOne thing is an individual recording their phone screen with another phone. Another thing entirely is having the operating system (OS) of your smartphone, or an agent AI of a specific application that you are obliged to use, that records everything all the time.
@veroandi Yes, that's Meredith's point. That having a developer level opt-out of these invasive agents would, in theory, make the OS prevent other apps from snooping out of their sandbox.
Then Okuna brought protection from other users bypassing the OS.
I think we're agreeing here.
@Okuna even if user is blocked to copy/paste or screen capture, nothing will prohibit to rewrite in agentic AI. Maybe copy/past is useful in other way, and disable it will reduce atractivity of the app. I understand your though. Some times ago I was depressed when I understood that my mom was sharing photos I sent to my self hosted Matrix/Element ... to WhatsApp, to see them in the google photo app timeline 😖.
Sometimes blocking to much things reduce adoptability in my case I had been treated as a dictador then I explain my right and give the two options element or signal... She stay on Element.
Sometimes blocking to much things reduce adoptability in my case I had been treated as a dictador then I explain my right and give the two options element or signal... She stay on Element.
Felipe 🐁@Mer__edith I would love this solution implemented, empowering the devs to choose.
Bjarne@Mer__edith Oh, you are right! Not sure why i didn’t thought about that the second they unveiled „Apple Intelligence“, showing it getting data from „Messages“ which is also E2E. Im wondering though. Apple mentioned it to fallback to ChatGPT when needed im not sure if this also applies to when Messages from „Messages“ will be sent. If so that would be crazy. (With that weird super secure apple server they had also showcased or not.)