Microsoft says AI agents pose insider-style security risks
Microsoft says companies should monitor AI agents as they would insider threats, especially agents employees use with outside systems. It also described new Microsoft 365 tools meant to contain that risk and adjust protections continuously.
By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .
This story is no longer being updated. It reflects reporting up to .
What happened
Microsoft has said that AI agents should be treated like insider threats, and it urged enterprises to watch closely the agents employees use with systems outside the control of AI labs such as OpenAI and Anthropic. The company described new capabilities in Microsoft 365 E7 and Project Perception that help organizations contain agent risk and adapt protection continuously. Microsoft's blog framed the concern against a background of publicly disclosed AI safety incidents.
Why it matters
Microsoft says powerful models can pose a significant threat if they are not handled well. Its guidance points to the need for organizations to keep agent risk contained.
Timeline
- Oct 19, 9:00 AM ET: Microsoft published a blog post on securing the agentic insider and announced new capabilities for Microsoft 365 E7 and Project Perception.
- Oct 19, 6:04 PM ET: Microsoft said that frontier models cannot be trusted and urged enterprises to closely watch the AI agents employees use with systems not controlled by AI labs.
Coverage
- Microsoft says AI agents are like insider threats as ‘frontier models can’t be trusted’ (techcrunch.com)
- Securing the agentic insider (blogs.microsoft.com)
- Agents are your newest insiders. Security must move at their speed. (microsoft.com)
- 4 ways to tame shadow AI (hint: you need them all) (computerworld.com)
More from Technology & AI
- Eight states sue OpenAI, alleging misleading AI safety claims
- OpenAI announces solution to Navier-Stokes math problem
- Datadog bets AI agents can handle testing of AI-written code
- Anthropic and OpenAI sketch wealth-sharing plans before IPOs
- Mamdani administration proposes AI customer service rules
- Florida engineers study how AI behaves in high-risk situations