Microsoft Unveils AI Agent Security Tools — SkimNews
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Microsoft introduced Guardian Agents, a new security feature under Project Perception, to monitor AI agents within enterprises and mitigate risks posed by unpredictable frontier models.
- Satya Nadella warned in a private blog that enterprises cannot outsource responsibility for AI actions, emphasizing that model providers' assurances do not absolve companies of oversight over agentic systems.
- Hayete Gallot, Microsoft’s security business chief, stated that frontier models 'cannot be trusted' due to their unpredictability and misalignment with organizational norms, not because of malice or vulnerability to hacking.
- Microsoft added agent detection to Microsoft Defender, capable of identifying up to 60 types of AI agents in the workplace, including Claude Code and OpenClaw, and now offers dynamic permission controls requiring human approval for high-risk agent actions.
- Microsoft joined Nvidia’s Open Agent Safety Platform, a coalition of over 100 tech firms, to promote isolated, hardware-enforced monitoring and kill-switch capabilities for AI agents.
- Microsoft launched Surface Laptop Ultra PCs designed for AI agents, powered by Nvidia chips and featuring built-in agent security controls, signaling deeper integration of AI safety into hardware.
Why it matters: Enterprises now face a shift in AI governance: trusting neither model providers nor their own deployed agents. Microsoft’s move creates a new security layer that could become standard, giving cloud providers like itself greater control over enterprise AI infrastructure while challenging rivals like OpenAI and Anthropic to prove their models can coexist with third-party oversight.
Ask SkimNews
