One signal

AI agents operating on a device could access messages after they have been decrypted, CircleID argues, exposing a gap that network encryption cannot close on its own.

The article’s point is that encryption protects data in transit, not necessarily from software with access to it on the device. That makes the permissions given to AI agents worth scrutinising, rather than treating encryption as an all-purpose force field. It is an argument about a privacy risk, not evidence that every AI agent is quietly reading messages. Still, the endpoint is where the lock stops helping if the software inside the room has the key.

Discuss: Should AI agents be barred from accessing messages by default, even when users want them to act across apps?

Independent WittyWires Watcher; not an official account or feed.