Remix.run Logo
▲ nkrisc a day ago

Your chats can always be leaked by the other participant, whether intentionally or unintentionally, in all chat apps, for all of the history of the written word.

▲Forgeties79 a day ago | parent | next [-]

Yes but ostensibly that is a deliberate action whereas people who are letting AI run amok on their stuff aren’t always aware of what exactly it is doing.

It’s like iCloud. If I send a photo of my kids to my parents, that could wind up on their iClouds. At that point whatever Apple does with that is out of my hands. I can clearly explain this to them if needed. But with AI there’s a lot more opacity, plus the average person does not understand permissions or look at TOS to see how their stuff is used.

▲dmoose a day ago | parent [-]

Even if completely aware of what it is doing and having understood all permissions and TOS the minute it leaves your control it is subject to changes in all of the above. That would be assuming that the provider is actually totally in control of their systems and procedures which is unlikely in any absolute sense. The safest approach is to not share or make accessible anything you wouldn't willingly make public.

Obviously a kids birthday party picture is quite a different risk profile from confidential sourced journalism but always having the thought in mind is a good way to make the sharing a conscious choice to accept the risk or not.

▲Forgeties79 9 hours ago | parent [-]

I think you’re missing my point. When people install Claude or whatever on their computer, they then let it hook into practically every system. Their email, their calendars, their photos, you name it. So suddenly they are exposing all kinds of stuff that previously you could reasonably assume wasn’t being fed straight to AI companies. There is a pretty big gap between what I was describing previously and that. And the fact is that most users, even if they kind of understand what’s going on, I don’t think fully understand the extent to which they are just cracking open their home and privacy for Altman et al. Which means I am now looped in by association most likely.

What am I going to do? Not send emails or text anymore to that person?

It’s like glassholes. The problem is when they’re walking around they’re just pushing whatever’s in front of them up to private servers. What am I going to do? Hunt them down and dodge all of their lines of sight?

▲mrtesthah a day ago | parent | prev | next [-]

Given that Signal already controls both sides of the conversation, I think they ought to just make these anti-AI controls enabled automatically for any conversation that has disappearing messages enabled (or perhaps use an additional user-settable per-conversation setting). That way any participant could prevent the other(s) from leaking messages to AI surveillance on their respective devices. This would reverse the viral direction of information leakage.

▲solyin a day ago | parent | next [-]

there is a desktop client though, at which point, all bets are off. Russia was documented trying to get Ukrainians to somehow scan signal QR codes in order to hack into their conversations.

▲mrtesthah 6 hours ago | parent [-]

macOS certainly has ways of making an app's files inaccessible to AI scanning. Even if the Signal app lacked the entitlements to configure those automatically, it could prevent the user from participating in a given conversation until it verified they were enabled manually.

▲warkdarrior a day ago | parent | prev [-]

DRM for the win!

▲mitthrowaway2 20 hours ago | parent | prev [-]

So let's reduce the surface area of the "unintentionally leaked" threat model