Fireworks AI gets burned twice as customer data exposure widens

Fireworks AI announced two separate cyberattacks in October that exposed customer data. On October 23, a third-party cloud provider hosting Fireworks' research and development machines was compromised, allowing attackers to gain access to Fireworks' development systems the next day. Fireworks suspects the attacks were financially motivated, possibly linked to LLMJacking – the sale of unauthorized access to AI models through illicit gateways.

The company categorized affected data into three groups: resource metadata, billing information, and model weights and datasets. Billing information included addresses, invoice histories, account balances, and credit usage. Payment card numbers were not saved and remained unaffected. Model weights and datasets for some customers were also accessed, though most downloads were incomplete and stopped.

Render CEO Anurag Goel reported that some Notion AI chat sessions may have been exposed. Notion confirmed that the exposed transcripts lacked personal user information. Fireworks is collaborating with customers like Cursor and Harvey to assess the impact. The company worked with Palo Alto Networks' Unit 42 team to remediate the attacks, which included rotating credentials, isolating affected systems, and tightening data storage access and monitoring.

Fireworks did not disclose the total number of affected customers. This incident follows an earlier disclosure on October 9, where unauthorized use of internal credentials was detected, but no evidence of model, dataset, or personal information access was found.