Microsoft unveils behavior-based project perception security controls for AI agents
Microsoft just dropped new security controls for its artificial intelligence, or AI, agents, aiming to keep things safer as these bots handle sensitive tasks like coding and data analysis.
Part of Project Perception, the update lets Microsoft limit what AI agents can access based on their behavior, so if something seems off, their permissions get dialed back automatically.
Microsoft executives stress adaptive AI security
AI agents need lots of data to do their jobs, but that also makes them targets for misuse or glitches.
Microsoft's Nadim Abdo, corporate vice president of identity and network access at Microsoft, said these new tools will help spot risks in real time and adapt as threats change.
Microsoft chairman and CEO Satya Nadella added that keeping security controls outside the AI models is key to stopping tampering.
It's all about balancing innovation with strong protection as AI gets smarter (and riskier).