Skip to content
AI.info

The Pulse

Reco reports eight Microsoft 365 takeovers, with AI role uncertain

An attacker used stolen sessions to take over eight Microsoft 365 accounts and create inbox rules that hid or deleted incoming mail, Reco reported. The company has moderate confidence the activity after access was automated, but low confidence that an AI agent was involved.

Reco reports eight Microsoft 365 takeovers, with AI role uncertain

An attacker used stolen sessions to take over eight Microsoft 365 accounts at one organization, creating rules that hid or deleted incoming mail, security company Reco reported. In the fastest case, a rule was in place 48 seconds after the attacker signed in.

The security company said it has moderate confidence that the activity after access was automated or scripted, with a human choosing targets. Its confidence that an AI agent drove that automation is low: conventional phishing kits and scripts could also explain the speed and consistency.

Two waves, one pattern

The activity came in two waves: three accounts were affected on October 23 and five between November 4 and November 9, 2026, according to Reco's November 18 account. The waves did not share an IP address, but the inbox rules had the same basic shape.

One account in the first wave was a less exact match. Its rule also marked mail as read, appeared hours after sign-in, and used infrastructure that did not overlap with the other accounts.

The rules had names of one to three characters and no conditions, so they applied to every incoming message. Each also had a setting called StopProcessingRules turned on, preventing later rules from acting on the mail.

Four rules deleted mail, three moved it to Archive and one permanently deleted it, the company reported. One rule was named with three periods; others used short names such as a, c and cc.

Actions before containment

On one account, the attacker generated 355 SharePoint sharing events in under three minutes, Reco said. The available data did not show who received access to the shared items.

On another, the attacker sent nine messages in about four minutes, with the last sent three seconds before the account was disabled. The recipients and subjects were not available, leaving the purpose of those messages unclear.

The organization's automated response disabled accounts and revoked refresh tokens about two to 15 minutes after the rules appeared; passwords were then reset, according to the report. Mail was read on one account 20 seconds after it was disabled.

How the sessions were obtained

Reco considers adversary-in-the-middle phishing a likely source of stolen sessions for at least three accounts. In that technique, a proxy relays a victim's password and multifactor authentication challenge to the real sign-in page, then retains the resulting session cookie.

On three accounts, Microsoft Entra ID raised an anomalous-token detection within about two minutes of the first sign-in, and the session was subsequently used from a second IP address, the company said. Its data did not show how the suspected phishing lures were delivered or how the attacker obtained sessions for the other accounts.

The company also observed a pattern of failed sign-ins affecting 4,725 identities at the organization. Reco said that activity looked like broad background password spraying and that it could not tie it to the account compromise.

Sources

  1. Original source · Blog post Reco reco.ai

Related stories

Explore

More articles