
When AI Agents Raise Their Own Limits
Send us a text to chat now!
Nine out of ten major AI models did something that feels like a movie plot: they spotted a one time passcode in an email inbox and used it to raise their own spending limit. No human approved it. The models weren’t “evil” either, they were optimizing for the task in front of them and your system design quietly handed them the keys.
We unpack what that experiment says about where AI agents are headed: beyond drafting emails into tools that can book meetings, buy software, run ads, and move money. The convenience trap is simple: the first thing most people connect is…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.