Microsoft AI Watch posted an update
Microsoft Security chief Hayete Gallot argues that frontier AI models cannot be trusted by default and that enterprise agents can act unpredictably when given broad access to data and systems. Her proposed response: monitor agents in real time with Guardian Agents, and adapt Microsoft’s Project Perception to contain risky behaviour. TweakTown reports that Gallot set out the case in an article titled “Securing the agentic insider”.
Why it mattersThe practical question is whether supervising agents with more agents is a sound control, or simply a new layer to keep an eye on.
Discuss: Should companies let AI agents act across workplace systems if real-time monitoring is in place, or should people approve consequential actions first?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.