

The speed part is what worries me most. When attackers can go from recon to exploit in hours instead of weeks, the old ‘patch it in the next maintenance window’ mindset just doesn’t hold up anymore. The boring basics matter more, not less: patch fast, especially internet-facing stuff, use phishing-resistant MFA, and actually look at your logs — most orgs only notice something’s wrong weeks later because nobody was watching for odd logins or weird outbound traffic.
And if you’re running AI agents or any automation with credentials, give them the least privilege you can get away with and keep thier access separate from human admin accounts. An agent that can read your whole codebase and deploy to prod is a great target, and if it gets comprimised through a prompt injection or a poisoned doc, the blast radius is huge. YMMV depending on your setup, but segmenting that stuff and logging everything the agent does is the cheapest insurance going.

Good move by Signal, and it’s worth understanding what this actually blocks. The feature stops the OS from letting another app grab the screen contents of Signal — that’s how on-device assistants like Gemini and Siri read what’s on screen when you ask them about it. It’s basically the same idea as FLAG_SECURE on Android, just applied to AI screen-reading instead of screenshots. It’s a seperate layer from the encryption — Signal’s messages are still end-to-end encrypted, this just covers what’s shown on screen.
A few caveats worth knowing though:
IMO this is the right direction — apps shouldn’t have to rely on the OS to keep AI assistants out of private chats, but untill the OS-level controls get better, per-app blocking like this is the practical answer.