Fireworks AI says an update to its October security incident has revealed additional customer data exposure after an intruder used a security breach at a third-party cloud provider to reach the company’s development machines. Account and billing information was affected, while secrets, datasets and incomplete model weights were accessed for a limited set of accounts.
In an update published on November 9, Fireworks said the third-party provider, which hosted its research and development machines, suffered a security incident on October 23. The unauthorized party used that access the following day to reach Fireworks development machines and run an unauthorized workload. The company said it detected the activity on October 24 and took remediation measures. Fireworks’ security bulletin provides the latest account of the incident, while Bloomberg’s original report described the customer notifications on November 9.
The affected internal application exposed resource metadata, including Fireworks resource identifiers, display names, descriptions and creation times, along with billing details such as billing and shipping addresses, invoice and payment history, account balances and credit usage. Fireworks says payment card numbers were not affected because it does not store them.
Fireworks said it has not detected further successful unauthorized access to the customer-facing platform after October 25. It also said it has not found an impact on Inference, Training SDK or Nexus, which use zero data retention by default. The company is notifying customers whose information was affected, and its investigation is ongoing.
The company said the evidence points to a financially motivated group seeking to sell tokens from unauthorized access to large language models through illicit routing gateways, a practice known as LLMJacking. Fireworks is working with Palo Alto Networks Unit 42 on forensic analysis. It said it has revoked and replaced credentials, isolated affected systems and tightened storage and monitoring controls. Self-service dataset uploads and downloads for Batch Inference and dataset-based training remain disabled while the investigation continues.



