Fireworks AI confirmed that a cyberattack the company suffered led to the leakage of some customer data. Fireworks AI is a platform that allows developers to access and use open AI models. The company stated that attackers infiltrated its cloud service provider, thereby gaining access to a development system of Fireworks and obtaining some customer data as a result.

In its latest security report, the company stated that cybercriminals stole billing information and portions of the data sets of Fireworks customers. Fireworks indicates that tens of thousands of customers around the world use its platform, including legal technology companies Harvey and AI, as well as Cursor.

It has been confirmed that affected customers currently include at least the online document editing and note-taking applications Notion. Notion stated in a disclosure that this incident affected approximately 1% of its customers.

Notion indicates that Fireworks was first notified of a security incident, followed by confirmation that hackers had stolen some customer data from Notion. Notion stated that on November 3rd, the company disconnected its systems from Fireworks, deleted the relevant data stored in their systems, and ceased all connections between the two parties.

Notion This week, it was confirmed that hackers obtained the chat records of Notion AI customers who use the “Insights” feature. The company stated that these records include “any model inputs or outputs written during the chats,” but also added that many of the data could not be attributed to a specific customer, and it was impossible to determine from which client the information was obtained.

The Fireworks AI leakage incident demonstrates that technology companies rely on other technology suppliers for their products and services, and as a result, cyberattacks may also spill over and affect other customers.

Fireworks suggests that the motivation of hackers is economic gain, with their focus on gaining access to large language models (LLM) in order to sell tokens. Such schemes rely on selling access capabilities to unauthorized third parties, allowing them to consume a large amount of usage of the AI model without having to use their own resources.

This is the second security incident that Fireworks has encountered within two months; the previous one occurred in October.