Fireworks AI Security Incident: What to Check If You Store API Keys on AI Platforms

What Fireworks says happened

Fireworks AI, a company that hosts and runs AI models for developers, published a security bulletin in October 2026. It says an unauthorised third party used internal Fireworks credentials to reach environment variables used in evaluation jobs, affecting a limited number of customer accounts. Fireworks believes API keys and tokens that customers stored as secrets were captured in some of those accounts.

According to the bulletin, the company identified the activity on 6 October, began notifying customers on 7 October and published the bulletin on 9 October. It lists a later update on 10 October, and the bulletin list we opened shows a further update dated 9 November, which we could not open.

What the company says was not affected

Fireworks says it has found no evidence so far of unauthorised access to customer models, datasets, training data, prompts and outputs, or personal information. It adds that it does not store prompts or outputs by default. It also says its review of calls made with the exposed credentials is still going on, so this is a current finding, not a final one.

What is not known

The bulletin does not say how many accounts were affected, how the internal credentials were obtained, or who the third party is. Fireworks says it has revoked the affected keys, added access restrictions and is working with an outside security firm.

RuntimeWire reported on 9 November, citing Bloomberg, that Fireworks had notified customers of a further intrusion dated 24 October that reached development machines and billing data, and for some accounts secrets, datasets and incomplete model weights. We could not confirm those details in the company bulletin we opened, and they do not match its timeline, so treat them as an unconfirmed report.

What affected customers are told to do

• Create new Fireworks API keys to replace revoked ones and update your applications.
• Rotate any third-party credentials named in the notice with the company that issued them. Deleting the stored secret does not invalidate the credential itself.
• Review access logs for those credentials, and rotate them even if nothing looks wrong.
• Look for unfamiliar API keys, users, evaluators, training jobs or datasets and report them to Fireworks.

Habits that limit the damage

Customers who did not receive a notice were not identified as affected so far, and Fireworks says no incident-specific action is currently required for them, though anyone who sees suspicious activity should make contact. The NHIMG write-up of the incident suggests wider habits: give evaluation and build jobs limited, short-lived, spend-capped keys instead of long-lived production keys, and keep a list of which platforms hold a copy of each AI provider key, so one notice can trigger a quick rotation.

These are good practice for anyone who connects AI services together; our guide to using AI at work safely covers related basics. We have not independently verified the incident, and details may change as Fireworks completes its review.