Microsoft announces Project Perception expansion to restrict risky AI actions
Technology
Microsoft just announced an expansion of Project Perception, its system for monitoring and managing AI agents.
The goal is to spot risks early and adjust what AIs can do based on how risky they seem, so if an AI starts acting up, its permissions get dialed back for safety.
Microsoft emphasizes oversight and alignment
Microsoft's move is all about not blindly trusting powerful tech.
Even well-meaning AI models can behave unpredictably or lead to "reward hacking," especially since they often have loads of access.
By treating AIs as complex team members who need oversight, Project Perception aims to keep them working safely and in line with human values.