Fireworks AI cyberattack leaks customer data, including Notion chat transcripts
Supply‑chain breach highlights interdependency of tech platforms
Fireworks AI, a developer platform that provides access to open AI models, disclosed that attackers compromised its cloud provider and then infiltrated a development environment. The intrusion granted the perpetrators access to data belonging to Fireworks’ clients, the company confirmed in a security advisory.
The hackers extracted billing details and several data sets from Fireworks’ systems. The platform serves tens of thousands of users globally, among them legal‑tech firm Harvey and AI‑focused company Cursor.
One of the identified victims is Notion, the online document and note‑taking service. Notion reported that the incident affected roughly 1 percent of its own user base. After receiving an alert from Fireworks, Notion severed the connection on November 3, removed the stored data, and halted all further integration.
Further investigation revealed that the attackers captured chat transcripts from Notion AI customers who used the “Insights” feature. According to Notion, each transcript contains every input and output exchanged with the model, though most entries could not be linked to a specific user.
The breach illustrates how a single compromised provider can ripple through the ecosystem, exposing data of downstream customers that never directly interacted with the attacker.
Fireworks characterized the intruders as financially motivated, suggesting they aim to monetize stolen access to large language models by selling usage tokens to unauthorized third parties. Such schemes allow buyers to consume model capacity without incurring their own costs.
This marks the second security incident at Fireworks in two consecutive months, following a separate breach reported in October.








