AI & agents

The emails and documents AI reads become part of access control

Published
Source
Microsoft Security Blog
Source type
Product announcement

Summary

Microsoft’s July security updates introduced detection of emails containing instructions designed to mislead AI, and controls to exclude external senders’ emails from Copilot’s reference material. Both were in preview at the time. Management is extending to what AI reads and uses to act.

Why it matters for our work

Text in news and external documents is material to read, not an instruction to our system. AI connections that retrieve external information need to preserve that distinction. Sources and read permissions should be recorded, while actions such as publishing or editing need a separate decision.

Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.

Read original (opens in a new tab)