Safeguard
Security Guides•

How do you cut off a compromised third-party app? A runbook

A step by step runbook for when a connected vendor app is breached: revoke tokens, review logs, rotate secrets stored in connected systems, and notify.

Safeguard Security Team
4 min read

Cut off a compromised third-party app by revoking its grants in every system it touches, reviewing its activity logs, rotating any secret that sat in those systems, and then blocking re-consent. Do containment first, investigation second, and write down times as you go.

Before the incident

You can only move fast if you know where the app is connected. Keep a register of connected apps with an owner and the revoke steps for each. If you do not have one, the first hour of an incident becomes a search. See the integration register template.

Runbook

StepActionNotes
1Name an incident lead and open a timelineRecord every action with a time
2Revoke the app's grants in each connected systemDelegated and application permissions are separate in some platforms
3Disable or delete the app's API keys and service accountsDo not wait for the vendor
4Block re-consentRevoking does not stop a user from granting access again
5Pull activity logs for the app's connection windowLook for bulk reads, exports, odd hours and new source addresses
6Search connected systems for stored secretsTickets, notes, files and messages often contain pasted keys
7Rotate every secret that could have been readStart with cloud keys and database credentials
8Decide who must be toldCustomers, regulators and partners, using your legal advice
9Ask the vendor for scope, dates and deletionGet answers in writing
10Review and restore only what is neededReconnect with narrower scopes

Why step 6 matters

In the August 2025 incident reported by Google Threat Intelligence, the actor searched the exported data for secrets such as cloud access keys and passwords. Google's recommended actions included searching connected platforms for such strings and revoking and rotating credentials found. Treat any shared system as a place where secrets may live, because people paste them into it.

Revoking correctly

Microsoft's guidance for Entra notes two cautions. Admin-granted and user-granted permissions are managed differently, and portal revocation covers the admin consent tab, so user consent grants may need an API or script. And revoking does not stop users from consenting again, so you must also change how users are allowed to consent.

Google advises developers to revoke tokens as soon as they are no longer needed. As a customer, revoke from your admin console and from the app's own settings.

After containment

  • Record what you learned and what slowed you down.
  • Reduce scopes on every app that survived.
  • Add detection: alerts for bulk exports and for new connected apps.
  • Run the revoke steps again in a drill, so the next time takes minutes.

Quick check

Should I wait for the vendor to confirm the breach?

No. Revoke on the first credible report. You can reconnect later. A delay gives an attacker more time with a working token.

Do I need to rotate secrets if logs show no misuse?

Logs can be incomplete. Rotate secrets that could have been read, starting with the most powerful.

Safeguard's asset discovery, policies and compliance evidence automation help you prepare the register and record the response. Try the free scan at gold.safeguard.sh/scan.

Sources

Last reviewed: 19 October 2026.

Never miss an update

Weekly insights on software supply chain security, delivered to your inbox.

Self-healing security runs on Safeguard.

Your first fix PR is minutes away.

No sales call required, even your agent can complete the purchase over MCP.