ROGUE PIXEL · Reported incident

Five minutes from scan to secrets.

The five-minute path is Microsoft’s observation in at least one case, not a general attack time. The final control check is proposed; Ackuity was not involved.

Read the narration

Microsoft says AI agents searched public GitHub for exposed secrets. In at least one case, an agent went from that scan to stealing key vault data in five minutes. Failed calls were retried with new API versions until they worked. Put the check outside the agent before each action.

Read the visual story

Public repository secrets lead toward a cloud vault. The five-minute observation is labelled as at least one case. Failed calls are retried with adjusted API versions. A final independent check illustrates a proposed safeguard.

Microsoft reports one case in which an AI agent moved from scanning public GitHub secrets to Azure Key Vault data theft in five minutes. Failed API calls were followed by adjusted requests.

THE CONTROL LESSON

What should we watch for?

Ackuity’s view: rotate exposed credentials and limit their reach. For agents you operate, independently check each action against the authorized task and target.

THE EVIDENCE

What does the report establish?

This is Microsoft’s account. Five minutes describes at least one observed scan-to-exfiltration path, not a benchmark or every campaign. HTTP 200 indicates a successful request, not an independent measure of all data taken. November 5 is the report date; scanning was observed from August and some follow-on sign-ins from October 28. The proposed check was not an Ackuity intervention.

Read the full account

Microsoft describes agentic secret scanning and subsequent credential abuse. It observed cloud requests fail, then succeed after HTTP-method or API-version changes.

Removing a leaked secret from a repository does not revoke it. Its permissions can still authorize further access.

Explore Ackuity’s Agent Execution Control Switch

Request an Agent Exposure Assessment