Microsoft Warns of 'Agentic Insiders' as AI Agents Reach Work PCs
Microsoft's security chief warns overprivileged AI agents are becoming 'agentic insiders' as on-device agents reach enterprise PCs — and governance, not the model, is the risk.

Priya Raman
· 3 min read
AI agents now run directly on Windows PCs inside enterprises, and Microsoft's security chief says most of them are overprivileged. On October 19, Hayete Gallot, Microsoft's Executive Vice President of Security, warned that organizations must do more to secure what she called "agentic insiders" — agents that can access sensitive data, tools and systems while making decisions in critical business processes.
The shift is no longer a forecast. NVIDIA's RTX-powered Windows PCs support locally running AI agents, and Perplexity has expanded its agent capabilities to compatible Windows computers, enabling multistep workflows that touch local files and connected applications. Microsoft, meanwhile, is building safeguards to control how agents access resources and take action. For HR and IT leaders, that means new people-process questions: who approves an agent's access to employee files, and who audits what it did.
What changed on the desktop?
Agents are moving beyond chat windows into the operating systems, applications and files employees use daily. Unlike a chatbot, an agent acts — it reads drives, edits files and triggers actions inside business apps. A wrong step can reveal sensitive information, alter critical files or set off unwanted activity in business applications, with repercussions beyond a simple wrong answer.
Microsoft's Windows security strategy rests on containment, limited permissions and limited oversight. Its agent-account and workspace model separates an agent's activity from a person's session, though the source notes those protections need to prove effective beyond demonstrations. Microsoft's Windows agent-security guidance calls for user authorization, activity logs and least-privilege access by design.
Does local processing mean private processing?
No — and that's the central governance warning. Local AI minimizes how much sensitive data moves to cloud services, but it doesn't guarantee privacy or security. An on-device model can still access too many files, run with poor security or hold unsafe links to other apps.
"The challenge is ensuring that an agent operates under the principle of least privilege, with access only to the data, systems, and actions required for a particular workflow," Bayron Quintero, CEO and co-founder of iData Global, told Unite AI. Quintero, whose company specializes in AI solutions across the Americas, said the problem usually isn't the model — it's organizational governance. Companies grant too much or too little access to data, applications and business processes, and agents end up with rights that don't match their tasks.
His verdict is blunt: there is no such thing as safe local AI. Excessive permissions, weak encryption, insufficient monitoring and insecure integrations can make an on-device agent as risky as a cloud-based one, or riskier.
What should employers require?
Quintero identified three essential safeguards:
- Clear data ownership, so someone accountable controls each dataset an agent touches.
- Role-based access controls matched to each specific workflow.
- Continuous monitoring of agent interactions with internal and third-party systems.
The permission logic is concrete. An agent asked to summarize one document shouldn't scan an entire drive. A photo-organizing tool shouldn't read financial data or send messages. Permissions must be tight, transparent and revocable — three tests any procurement or IT policy can apply before an agent reaches employee desktops.
Where's the next risk?
Prompt injection through untrusted content. During an authorized task, agents encounter documents, websites and messages they didn't choose, and Microsoft's guidance tells employers and developers to monitor how agents react to misguided instructions embedded in that content. When an agent treats hostile text as a command, the exposed information can be easily accessed. Human approval and least-privilege access must go hand-in-hand with containment for sensitive actions.
The AI PC will ultimately be judged not on how much work its agents automate, but on whether users trust it with personal and professional data. As Quintero and Microsoft's own guidance make clear, control can't be an afterthought — it has to sit inside the design as computers become more capable of acting for us.
Source: unite.ai
More from Priya Raman
Show full bio
Staff writer covering companies, deals and strategy in Recruiting and talent acquisition, learning and development at HR Briefing.
273 articles