Fireworks discloses intrusion into evaluation jobs, with customer keys potentially captured
The company says it found no evidence of access to models or inference traffic within its control, but affected customers must also check credentials issued by other providers.
Listen to this story
The audio brief
Story brief
3 key pointsFireworks says an intruder accessed environment-variable settings used by evaluation jobs for a limited number of customer accounts, and that those settings likely contained API keys and other provider tokens. As of October 10, it had found no evidence within its security perimeter that customer models, datasets, training data, personal information, or inference traffic were accessed, but its review of credential use remained open. Affected customers need to replace revoked Fireworks keys and rotate third-party credentials; deleting stored secrets alone does not invalidate them.
- 01
Fireworks identified unauthorized use of internal credentials on October 6 and listed the incident as under active monitoring in its October 10 update.
- 02
The company revoked identified internal and affected customer keys, added access restrictions, and continued reviewing how evaluation jobs receive secrets.
- 03
Affected customers were advised to review provider access logs and check for unfamiliar keys, users, evaluators, training jobs, or datasets.
Fireworks says an unauthorized third party accessed settings used in evaluation jobs for a limited number of customer accounts. It believes those settings included captured API keys and tokens stored as secrets. Its security bulletin directs affected customers to replace revoked Fireworks keys and rotate exposed credentials issued by other providers.
Exposed secrets, an unfinished access review
The accessed settings were environment variables: values supplied to the evaluation jobs, including credentials stored as secrets. Fireworks identified unauthorized use of its internal credentials on October 6, 2026. The company describes the affected group as a limited number of customer accounts, without giving a count.
As of its October 10 update, Fireworks said it had not detected evidence of unauthorized access to customer models, datasets, training data, personal information or inference traffic, including prompts and outputs. That finding applies within Fireworks’ security perimeter or control. Its review of calls made with exposed credentials, and of other potential access, remained ongoing.
Fireworks says its platform does not store prompts or outputs by default. It is also working with a cybersecurity expert firm on the incident. The bulletin was last updated at 11:16 a.m. PDT on October 10 and listed its status as active monitoring.
Containment includes how jobs receive secrets
Fireworks revoked the identified internal credentials used in the incident, affected Fireworks API keys and keys identified as created by the intruder. It also added access restrictions. Further credential review and rotation were continuing at the time of the update.
The company is reviewing how secrets are made available to evaluation jobs. It says it will share additional remediation measures once they have been implemented and verified.
Deleting a stored secret is not enough
Affected customers received emails identifying credentials requiring action. Fireworks’ customer guidance distinguishes between keys it has revoked and third-party credentials that customers must address with the providers that issued them. Removing a secret from storage does not invalidate the underlying credential.
- Replace revoked Fireworks keys still needed by applications. Create new keys in the Fireworks account and update those applications; they will fail authentication until their revoked keys are replaced.
- Rotate or revoke third-party credentials named in the notice through their issuing providers. Then update the applications and Fireworks secrets that use them, even if no suspicious activity is found.
- Work with those providers to review access logs and data reachable with the exposed credentials. Look for unauthorized reads, downloads or changes, and follow provider guidance if unauthorized access is found.
- Check API keys, users, evaluators, training jobs and datasets for unfamiliar resources or activity. Send names or IDs to Fireworks so it can investigate whether they are connected to the incident.
Fireworks can supply secret names and credential fingerprints through a secure channel to help customers identify exactly what needs rotation. Customers can reply to their incident notification or contact security@fireworks.ai.
For customers who have not received a notice, the bulletin says no incident-specific action is required based on the investigation to date. Fireworks says it will contact customers directly if their status changes, and asks anyone seeing suspicious activity to get in touch.
Sources
- fireworks.aiFireworks October 2026 security incident
Reader comments
Newest comments first. Replies stay oldest first.