Fireworks AI gets burned twice as customer data exposure widens

What the attack exposed

Fireworks AI disclosed that the October incident involved resource metadata, billing information, and limited access to model weights and datasets for certain customers. The metadata included resource identifiers, display names, descriptions, and creation times.

Billing records may include addresses, invoices, payment histories, account balances, and credit usage. Fireworks said payment card numbers were not stored and were unaffected. The company did not specify how many customers were affected by either October attack.

Suspected motive and response

Fireworks suspects the attacks may be linked to financially motivated criminals involved in LLMJacking, which involves selling unauthorized access to AI models through illicit gateways. Co-founder and CEO Lin Qiao said the apparent motive was not theft of model weights, addressing concerns raised by Cantina Security CEO Hari Mulackal.

Notion chats may be involved

The notification said the chat transcript was the only information potentially exposed, with no details linking it to a workspace, user, or company. Notion also said no other workspace data was affected.

Earlier assessment revised

Fireworks had previously said it found no evidence that models, datasets, training data, inference traffic, or personal information were accessed. Monday’s update substantially expands the disclosed impact to include billing records and some incomplete model weights and datasets for certain customers.