It’s weird they haven’t just made this a menu flag…
I'm sure this one probably isn't, but I have no clue how I would know whether it is without reviewing the source code myself (or pointing an agent at it etc)
Do you think a malicious repo author won't think of hiding a suitable prompt within?
(not saying this is ai or not, i have no idea, just speaking generally)
We're getting free from our software!
I know it's well intentioned but I have this weird setup where I am almost constantly keeping vnc on from my laptop to my desktop. And the mbpro randomly wakes up to do whatever, even when not used, which reactivates the vnc connection, which pops up a warning on my desktop, which steals my focus.
Anyone knows a way to disable that? Possibly just for connections from a specific other machine.
Edit: as for this app, I think the OS only asks once for a new app? I don't remember being asked a second time even when denying it.
I could also whine a bit about "this app wants to access files on a removable volume" since my desktop has a thunderbolt ssd plugged in for extra space...
You can see other people with tens of instances of the same app in the settings for this
Same with MDM agents. Ours are installed automatically as part of the Apple Business Manager configuration, we manually grant them precise location in the background, and months later "HEY USER, WANNA BLOCK LOCATION FROM 'MDM SERVICE'?".
Show me just one where a frontier model got prompt injected when checking if code if malicious.
FWIW I can do this, but I bet you can't.
Here's how it was done via a harness: https://www.promptarmor.com/resources/google-antigravity-exf...
took me about 30 seconds to find. there's tons of them. of course it's an arms race, the models get better the attackers get better but again fundamentally: the model itself does NOT distinguish between your prompt and data it reads.
poszlem•1h ago
klodolph•58m ago
johncassil•34m ago