Fireworks AI, an artificial intelligence infrastructure provider, has confirmed a cyberattack that exposed customer billing information, datasets, and AI-related data, including chat transcripts belonging to Notion users. Notion disclosed that fewer than 1% of its workspaces were affected after attackers gained access to Fireworks’ development systems through a third-party cloud provider.
Fireworks provides infrastructure that allows developers and businesses to run open-weight AI models, serving tens of thousands of customers worldwide, including Cursor and Harvey. The company said attackers accessed an internal system containing customer information following a security incident at one of its cloud providers.
Attackers Access Customer Data Through Cloud Provider
According to Fireworks’ security advisory, a third-party cloud provider hosting its research and development systems experienced a breach on October 23. The following day, attackers used that access to compromise Fireworks’ development machines and execute an unauthorized workload.
The compromised systems exposed resource metadata, billing and shipping addresses, invoice histories, account balances, and credit usage information. Some customers also had portions of their datasets and incomplete AI model weights accessed, although Fireworks said payment card numbers were not affected.
Fireworks detected the unauthorized activity on October 24 and implemented containment measures. The company said it had found no further evidence of successful unauthorized access after October 25.
Notion Confirms AI Chat Transcript Exposure
Notion said Fireworks initially notified it on November 2 that attackers had accessed a dataset containing synthetic testing information. By November 3, Notion had disconnected its systems from Fireworks, deleted stored data, and stopped communications between the platforms.
On November 8, Fireworks confirmed that additional customer data had been compromised, including Notion AI chat transcripts associated with its Insights feature. The transcripts contained user inputs and AI-generated responses, although most lacked identifying information connecting them to specific users or workspaces.
Notion said fewer than 1% of its workspaces were potentially affected and that no other Notion features were compromised. The company disabled Insights, stopped using Fireworks to serve open-weight models, and advised affected customers to replace any passwords, API keys, or credentials included in exposed conversations.
Second Security Incident Linked to AI Access Theft
Fireworks believes the attackers were financially motivated and involved in a practice known as LLMJacking, which involves stealing access credentials to sell unauthorized AI model usage to third parties. The company had previously disclosed another security incident in early October involving compromised stored secrets, including API keys and tokens.
Fireworks has engaged cybersecurity firm Palo Alto Networks Unit 42 to assist with its investigation and has introduced additional access restrictions, network isolation measures, and monitoring controls. The company also temporarily disabled certain dataset upload and download functions while continuing to investigate the breach.
Featured image credits: Magnific.com
For more stories like it, click the +Follow button at the top of this page to follow us.