Is it possible to limit an AI's interactions with the Internet?

Once an AGI has access to the internet it would be very challenging to meaningfully restrict it from interacting with the internet however it wants to. There are too many ways to bypass blocks we may put in place.

It may be possible to set up "tripwires" to that alert us when an AI is trying to do something dangerous online, though the effectiveness of this method relies on the AI not noticing or bypassing the tripwires.

Alternate phrasing