Microsoft Agent 365

Headlines · 2

1. 3d
   Nadella says AI alignment will not emerge automatically from scaling

   Microsoft CEO Satya Nadella said he no longer believes that making models more capable will automatically produce AI alignment, calling it a harder and different problem. He argued that alignment must largely occur during training, possibly from pretraining, while urging companies to use zero-trust security, tamper-proof logs and audits around AI models and agents. Nadella also said chain-of-thought can be unfaithful and that Microsoft will sell its MAI models as licensed weights rather than release them as open-weight models.

   The Next Web · 🔥 0
2. Oct 19
   Microsoft unveils Guardian Agents to police AI agents as insider threats

   Microsoft on Monday announced new security features in Microsoft 365 E7 and its Project Perception platform that treat AI agents as potential insider threats, with Guardian Agents as the centerpiece: another set of AI agents that check other agents' actions, investigate anomalies and apply limits. Microsoft Security chief Hayete Gallot wrote that frontier models, regardless of origin, cannot be trusted because they are unpredictable, while CEO Satya Nadella argued companies must separate the supply of intelligence from authority over it. The rollout also includes dynamic permissions that give an agent only the access one task needs, dynamic risk scoring, conditional access that can restrict or stop an agent, and Defender discovery of agent types such as Claude Code, Codex and GitHub Copilot, with reports differing on whether the total is 60 types or 35 additional ones. Microsoft did not say when the features will be available or whether they cost extra beyond an E7 licence.

   iThome 台湾 · 🔥 0

Experience and discussion from the community

Share my Microsoft Agent 365 experienceAsk about Microsoft Agent 365