On 5 November 2026, Microsoft Threat Intelligence published a report that should change how every company treats a leaked password or API key. Since August 2026, attackers have been using agentic AI penetration testing frameworks, including one called ARTEX, to search public GitHub repositories for exposed credentials and then use them to break into cloud environments. In one case Microsoft observed, an AI agent went from searching GitHub for secrets to exfiltrating data from Azure Key Vault in five minutes.

Our view is simple. If a secret has ever been committed to a public repository, deleting it from the code is not a fix. The only fix is to revoke or rotate it, and the time you have to do that is now measured in minutes, not maintenance windows.

What Microsoft actually found

The attacks start with large automated searches on GitHub. According to Microsoft, the queries combined the name of a target organisation with keywords for secrets, a keyword linked to cloud services, and in some cases a keyword linked to an AI service provider. Once the tooling finds a valid application secret, it signs in as that application and carries on from there.

Four-step attack chain: automated GitHub search, secret found, service principal sign-in, data exfiltrated from Key Vault and Storage Accounts in about five minutes in one case

What changes with agentic tooling is how little a human has to do. Microsoft notes that these frameworks let attackers scan many repositories, pick out the valuable credentials, and run follow-on actions, cutting down on hands-on-keyboard work across the attack chain. From 28 October 2026, Microsoft began seeing successful service principal sign-ins involving multiple application IDs across affected customers. The activity included credential and data exfiltration from Azure Key Vaults and Storage Accounts, access to SharePoint and OneDrive, and abuse of Azure AI services.

Why deleting the line does nothing

When developers discover a key in their code, the instinctive move is to remove the line and push a new commit. The repository looks clean. The key is still valid.

Microsoft's guidance is blunt: treat credentials that have been publicly exposed as compromised, even if the original location has since been edited or deleted. GitHub's own documentation explains why. If you only rewrite history and force push, commits containing the sensitive data can still be reached in clones and forks of the repository, and directly through their SHA-1 hashes in cached views on GitHub. Microsoft adds edit history, caches, archives and logs in other systems to that list.

Where a deleted secret can still be found: clones and forks, GitHub caches and archives, attacker scan results; core message is revoke first, clean up second

There is one more copy nobody can delete: whatever an automated scanner already recorded. An agent that found your key at 09:00 does not forget it because you removed the commit at 10:00. That is why GitHub puts revoking or rotating the secret as the first step, before any history rewriting.

The risk you did not create

The most uncomfortable detail in the report is about vendors. Microsoft describes a case where a multitenant application credential was compromised at the source, in a vendor-owned application, and then used to reach multiple downstream customer tenants.

Hub diagram: one compromised vendor-owned multitenant application connected to customer tenants, Key Vault, Storage Accounts, SharePoint and OneDrive, and Azure AI services

For many Indonesian businesses this is the realistic scenario. The internal system, the e-commerce integration, or the reporting dashboard was built by an outside developer or agency. That vendor holds cloud credentials, API keys and database access for your environment, and possibly for dozens of other clients. Your team never committed anything. If the vendor's key leaks, you are exposed anyway.

So the question for business owners is not "is our code clean?" but "which keys does our vendor hold for our systems, and when were they last rotated?" A healthy answer is a list and a date. "It's secure" is not an answer.

What to do this week

Microsoft's recommendations translate into a short list most teams can start on immediately:

  1. Scan your repositories and their history for secrets, including old branches and forks you control.
  2. Revoke or rotate anything that was ever exposed, then review sign-in logs, Key Vault and Storage diagnostics, and audit logs for use of the old credential.
  3. Turn on secret scanning and push protection in GitHub so supported secrets are blocked before they reach the repository.
  4. Move applications to managed identities where possible, so there is no secret written in code to find.
  5. Apply least privilege to service principals and other workload identities, so one leaked key opens as little as possible.
  6. Ask your vendors for an inventory of the credentials they hold and their rotation dates.
Comparison of three responses: deleting from code leaves the key working, revoking and rotating disables it, managed identity removes the secret from code entirely

The real shift

Leaked keys are not new. What is new is that finding and using them has become cheap and fast. When the attacker's side is automated, a rotation process that waits for a ticket, a weekly meeting, or a vendor's reply is already too slow. Companies that rotate on discovery, and that move away from long-lived secrets in code, will absorb the next leak as a minor incident. Those that still treat "deleted from the repo" as "done" are leaving the door open for whoever scans next.

If you are not sure which keys are spread across your repositories or held by your vendors, XETUP can help you audit your code and cloud accounts, find secrets that were committed, and coordinate rotation with your vendors. Reach us at xetup.id, bisnis@xetup.id, or WhatsApp +62 822 6515 7644.