What happened
Between June 1 and June 13, 2026, some EMU accounts deprovisioned through SCIM were deleted. They should have been suspended and obfuscated. The visible result was that affected users showed up as "ghost users" in issues, pull requests and audit logs. They did not keep their expected suspended, obfuscated records inside the enterprise.
If you have ever opened an old pull request and found the author replaced by a generic ghost avatar, you know the problem. It breaks the "who did this, and when?" chain that audits depend on.
How many accounts were affected
GitHub's figures:
- About 12,285 EMU accounts across multiple enterprises were affected.
- Most of them had already been suspended and hard-deprovisioned, so they couldn't be reprovisioned by design.
- A smaller group of 76 accounts were active when they were deleted.
The notice doesn't name any enterprises or give a per-enterprise breakdown. It also doesn't describe the status of any accounts beyond those two groups. Don't read "76" as the only active accounts affected, or as the total number of people who lost something meaningful.
Cause and containment
On June 1, 2026, GitHub enabled a change that sent deprovisioning requests down a new path. That path didn't correctly check whether the account was an EMU account. As a result, accounts were deleted instead of following the suspend-and-obfuscate flow. GitHub contained the problem by disabling that path on June 13, 2026.
This is a familiar failure pattern. A new code path skips a guard that the old path enforced, and the missing check is the whole bug. GitHub gives no further technical detail. The changelog is also the only account of the incident, so the cause, scale and repair status all rest on GitHub's word.
Why deletion contradicts normal EMU behavior
GitHub's SCIM documentation says that a deprovisioned account is suspended, and regardless of the type of deprovisioning it is never deleted from an enterprise. That makes the incident a departure from documented behavior. It isn't a case of hard-deprovisioning working as intended.
The docs also separate two kinds of deprovisioning:
- Soft-deprovision: in certain scenarios, the user can be unsuspended through the SCIM integration.
- Hard-deprovision: which kind of call GitHub receives from your IdP determines whether reinstatement is possible. Per the documentation, a hard-deprovisioned user can't be unsuspended, and a new account has to be provisioned if the person needs access again.
For hard deprovisioning, the documentation says resources the user created, such as comments, are retained. That is why the ghost-user symptom matters. The history was supposed to remain attributed to an obfuscated enterprise record. The bug broke that link.
How to check if your enterprise was affected
GitHub's test is simple:
- Open your enterprise audit log.
- Look for a
user.deleteevent for an EMU account between June 1 and June 13, 2026. - If you find one, your enterprise was affected. If you find none in that window, it wasn't.
GitHub doesn't provide a query, UI path or API call for this check, so use whatever audit log search or streaming tooling you already have. The SCIM docs strongly recommend that owners enable audit log streaming. That lets you set your own retention policy and use your preferred tools to query the logs.
What admins need to do
Nothing, according to GitHub. If you see those events, no action is required. GitHub says it restored the affected user records in October 2026. The restoration returned the accounts to their expected suspended and obfuscated state, preserving traceability and audit history within the enterprise.
Some boundaries on what that means:
- Restoration doesn't reactivate anyone. Hard-deprovisioned accounts can't be reprovisioned by design, and GitHub says the reprovisioning workflow isn't affected.
- The notice doesn't report lost repository content or unauthorized access. It also doesn't ask you to change your IdP or SCIM configuration.
- The changelog contains one oddity. It says "GitHub is restoring the affected accounts" in one place and that the records were restored in October in another. That is probably wording that wasn't updated between drafts. If your audit trail matters for compliance, it is worth confirming your own records through GitHub support rather than assuming.
Takeaways for IT and security teams
- Compare against your own records. If a compliance review or incident investigation covers June 2026, note that attribution in that window may have looked wrong for a time. Keep a note in your audit documentation.
- Stream your logs. Enterprise audit logs are your independent evidence. Streaming them means you aren't relying on GitHub's retention or its restoration.
- Know your deprovisioning type. The docs say Okta sends only soft-deprovisioning calls. Entra ID hard-deprovisions when the Entra user is hard-deleted, and a REST API DELETE request hard-deprovisions too. Knowing which one your IdP sends tells you what recovery options you have.
- Expect delays in vendor disclosure. The incident was contained in June, and the public notice and restoration came in October. That is more than four months. It is reasonable to ask vendors about their notification timelines.
Bottom line
This was a contained, retrospective bug. It caused no announced loss of access control and no data exposure. It did damage the attribution and audit history that identity governance relies on. GitHub says it has fixed the data and the code, but its own changelog is the only evidence available. If you manage EMU, a quick audit log search for user.delete events in that window takes a few minutes and is worth doing.
References
- EMU deletions fix GitHub Changelog · 2026-10-23T18:56:18+00:00
- Deprovisioning and reinstating users with SCIM - GitHub Enterprise Cloud Docs docs.github.com
- Deprovisioning and reinstating users with SCIM - GitHub Enterprise Cloud Docs docs.github.com