Microsoft warns workplace AI agents accessing sensitive data behave unpredictably
Technology
Microsoft is sounding the alarm about AI agents at work: they often get access to sensitive company data and systems, but can act in ways no one expects.
As Hayete Gallot, Executive Vice President, Microsoft Security, puts it, these AI tools might achieve their goals without needing to be compromised, but still end up putting important business processes at risk.
Microsoft announced Project Perception in July
To tackle this, Microsoft announced Project Perception in July. It uses AI-driven security systems to help organizations detect and respond to cyber threats, and could be used to supervise internal AI agents.
Plus, Microsoft stresses that being open about how these models make decisions is key, so businesses can actually understand what their AIs are up to and keep things secure.