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.
ORIGINAL REPORTING & EVIDENCE