Fireworks AI Faces a Second, More Serious Reported Hack

By Tom Whitfield

Fireworks AI raised $1.5 billion in July at a $17.5 billion valuation, according to MLQ. The Information then reported in a Monday article citing two people with knowledge that Fireworks had told large customers, including legal-AI company Harvey, about a more serious hack during the past week and that the intrusion exposed data from some customers.

MLQ could not independently confirm the incident at its Nov. 9 check. The Information article was paywalled, Fireworks' public security bulletin had not been updated since Oct. 10, and neither Fireworks nor Harvey had commented publicly. The startup runs and fine-tunes open-source models for developers.

The chronology therefore divides into two categories: a publicly detailed October incident and a newer, externally reported breach. Fireworks had not published the technical scope of the second event.

What did Fireworks disclose in October?

Fireworks disclosed its first breach on Oct. 9. It said an unauthorized third party had used internal Fireworks credentials to access environment variables in evaluation jobs for “a limited number of customer accounts.” Those variables could hold API keys and tokens that customers had stored as secrets, so Fireworks said those credentials were likely captured.

The company said it had found no evidence of access to customer models, datasets, training data, personal information, or prompts and outputs. It also said it does not store prompts or outputs by default.

Fireworks detected the activity on Oct. 6 and notified affected customers from Oct. 7 through Oct. 9. It revoked exposed Fireworks API keys and instructed customers to rotate third-party credentials with the providers that issued them.

Takeaway: In plain terms, the October disclosure described exposure of stored secrets, while Fireworks reported no evidence that customer models or training data had been accessed.

What does the newer breach change?

The Information described the newer incident as more serious than the October event and said it exposed customer data. The portion available to MLQ did not identify the type of data involved or say how many customers received notice. It also did not establish whether the October findings about models, datasets, personal information and prompts apply to the incident.

The report named Harvey among the large customers notified, but did not say whether Harvey's data was exposed. Notification alerts a customer that its systems or credentials may be at risk; it does not, by itself, prove that the customer's models or records were taken.

Fireworks' instruction to rotate third-party credentials with their issuers shows that the October response did not stop with Fireworks-issued API keys. The newer report now concerns unspecified customer data, but it supplies no technical basis for deciding whether the earlier model and dataset assurances still apply.

Takeaway: The newer report raises the risk profile, but the available evidence does not define the exposure.

What is the commercial exposure?

The Information's headline described Fireworks as a $1 billion business. Index Ventures said Fireworks' annualized revenue had “passed $1 billion.” Fireworks recently hired Meta strategy chief Dave Wehner as its chief financial officer, as MLQ reported.

The financing and revenue figures establish Fireworks' scale, but neither measures the likely cost of the breach. The report supplied no loss estimate, affected-customer count, service-interruption period or remediation expense. The commercial test will be whether large customers keep evaluation and fine-tuning workloads on the platform after receiving notice of a second breach.

For a platform whose annualized revenue had passed $1 billion, a broader customer-data exposure could weaken pricing power and customer retention if large accounts move inference and fine-tuning workloads to rivals.