Kyvvu

Agent incident database

EchoLeak: one email could make Microsoft 365 Copilot leak internal data

11 Jun 2025Microsoft · Microsoft 365 Copilotdata exfiltrationdisclosed vulnerability

Researchers at Aim Security showed that a single email, never opened by the victim, could cause Microsoft 365 Copilot to send internal company data to an attacker. Microsoft rated it critical and fixed it on the server side.

How it unfolded

Each step on its own looked like ordinary agent work. The harm came from the sequence.

  1. 1

    Received an email from an outside sender containing hidden instructions

  2. 2

    Retrieved that email together with internal documents to answer a user's question

  3. 3

    Model followed the hidden instructions and wrote internal data into an image URL

  4. 4

    The image was loaded automatically via a trusted Microsoft domain, sending the data to the attacker

    harm

Which policy would have stopped this?

In our reading of the reports, these policies from the open library stop the harmful step before it runs.

Open External content taint: high-impact actions require a fresh gate in the playground and build the path yourself.

This is our reading of public reporting. We haven't run it against the original systems.

What happened

Aim Labs reported the flaw to Microsoft in January 2025 and published it on 11 June 2025 as EchoLeak (CVE-2025-32711, CVSS 9.3). The attack starts with an email whose hidden instructions are phrased as if addressed to a human, which let them slip past Microsoft's cross-prompt-injection (XPIA) classifier. When the user later asks Copilot a related business question, its retrieval engine pulls the email into the model's context alongside sensitive internal data. The injected instructions get the model to place that data into a markdown link or image URL, using formatting tricks that avoided Copilot's link redaction. Because the browser's content security policy trusted Microsoft Teams and SharePoint domains, those URLs could be used to carry the data to the attacker's server automatically, with no click.

Outcome

Microsoft fixed the issue server-side in May 2025 and listed it in its June 2025 Patch Tuesday; no customer action was required. Microsoft said it found no evidence of real-world exploitation.

Sources

A summary of public reporting. The step-by-step timeline is our reading of those sources; where they differ, we say so above.

Your agents

Could this happen to yours?

For security teams

Request a demo

We walk through your agents' riskiest paths with you.

For developers

Build a path yourself

Recreate a sequence in the playground and see how policies treat it.