An exposed AI conversation can create a security problem long after the chat ends.
Notion said in a Nov. 13 update that it reported exposed third-party API keys to their issuers on Nov. 12, following its Nov. 11 revocation of exposed Notion keys. The credentials appeared in user messages affected by the Fireworks AI incident.
For security teams, the next question is practical: Which services could any still-active credentials unlock?
What changed in Notion’s response
The latest Notion advisory adds credential-specific actions to the incident already covered in eSecurity Planet’s Fireworks breach report.
Notion says fewer than 1% of its workspaces were affected. The exposed dataset included AI chat transcripts used for its Insights feature. The company says its own systems were not breached.
Affected customers can securely review the impacted data in their workspaces using the emailed instructions. Notion recommends treating that material as exposed and rotating sensitive credentials pasted into affected conversations.
Customers who have not heard from Notion have not been identified as affected and need no action right now, the company says. Its investigation continues.
How the Fireworks incident widened
In its Nov. 9 bulletin, Fireworks described additional activity after an earlier incident disclosed in October.
The company said a third-party cloud provider hosting its research and development machines suffered a security incident on Oct. 23. The following day, an unauthorized party accessed Fireworks' development machines and ran a workload that caused further impact on customer data.
Fireworks said it detected that activity on Oct. 24 and took remediation measures. In its Nov. 9 update, it reported no further detected successful unauthorized access to its customer platform since Oct. 25, while emphasizing that its investigation remained ongoing.
Its response included replacing credentials and service accounts, removing unnecessary elevated permissions, restricting network access, and strengthening logging. Palo Alto Networks Unit 42 is assisting the investigation.
That wider timeline matters for affected businesses: A provider’s containment milestone does not answer whether a credential exposed earlier remains usable at another service. The latest Notion update makes that distinction especially important for teams managing integrations across several platforms.
Why an API key needs a separate response
An API key identifies an application or authorizes its requests to a service, depending on how the service uses it. Possession of a key may allow someone to make requests or access resources without another interactive login.
That distinction matters when investigating an AI conversation containing technical configuration, troubleshooting notes, or copied code. The sensitive item may belong to a completely different platform.
As eSecurity Planet’s API security overview explains, systems that treat a key as an authoritative credential can be vulnerable to anyone who obtains it. Security teams therefore need to assess what each exposed key authorizes, rather than treating all credentials as equivalent.
Notification is also different from confirmed revocation. Notion’s update does not establish that every exposed third-party key has been revoked. Administrators should verify the status of affected credentials with the issuing service rather than assuming that a vendor notification has completed cleanup.
What affected teams should check
Once a workspace is identified as affected, assign the review to someone who can coordinate with application owners and security staff.
The following checklist applies general OWASP secrets guidance to credential cleanup:
- Identify the service and owner. Record which application uses each credential, what it can access, and who is responsible for replacing it.
- Revoke exposed credentials. Confirm that the issuing service has disabled the old secret. Removing a copy from a conversation is not a substitute.
- Deploy replacements carefully. Update dependent applications and test them so credential changes do not leave integrations broken.
- Review relevant activity. Examine available access records for unexpected use, including attempts involving revoked credentials.
- Keep an incident record. Document what was exposed, when it was disabled, and which systems were checked. Avoid copying live secrets into the incident ticket.
These are investigation and containment steps, not evidence that every exposed key was used maliciously.
Check the systems beyond the AI tool
A useful closeout question is whether the old credential still works anywhere it should not.
The broader lesson also appears in eSecurity Planet’s Zeabur breach coverage: Recovering one platform does not automatically revoke access granted by a credential elsewhere.
For an affected team, acknowledging the notification is only the start. The next steps are to account for relevant credentials, test replacement integrations, and investigate suspicious activity.
More cybersecurity coverage: For another example of how compromised credentials can reach beyond their original platform, read about the OpenCode API-key breach.





