Fireworks AI Breach Exposes Customer Data and Partial Model Weights via Cloud Provider

Fireworks AI expanded its breach disclosure on November 9, confirming exposure of customer account information, billing records, datasets, and portions of certain customers’ incomplete model weights. An October 23 security incident at an unnamed third-party cloud provider enabled attackers to access Fireworks development machines and run an unauthorized workload on October 24. The findings broaden the exposure beyond credentials reported in October; an earlier incident detected October 6 involved misuse of internal credentials and access to evaluation-job environment variables that potentially exposed API keys and tokens.

Fireworks said evidence points to financially motivated criminals seeking to resell unauthorized large language model access through illicit routing gateways, a practice known as LLMJacking. The company reported no detected impact to Inference, Training SDK, or Nexus, and the disclosure does not establish theft of complete production models. Fireworks engaged Palo Alto Networks Unit 42, notified law enforcement, and restricted affected functionality, including some dataset operations. Affected customers should follow the company’s guidance to rotate exposed credentials and review account activity while the investigation continues.