Fireworks AI Breach Exposes Notion AI Users’ Chat Transcripts

A cyberattack on Fireworks AI has exposed the chat transcripts of Notion’s AI users, underscoring how a single breach can spill over into other companies. Attackers breached Fireworks’ cloud provider, gaining access to a development system that stored customer data. The breach stole billing information and some datasets, including chat logs from Notion’s “Insights” feature.

Much of the stolen data is unattributable, making it hard to identify which customer’s information was taken. Fireworks serves tens of thousands of customers worldwide, from legal tech firm Harvey to AI startup Cursor. Notion, one of the affected clients, said the breach hit about 1% of its own customers.

In response, Notion disconnected its systems from Fireworks on November 3, deleted the stored data, and cut all connectivity between the two services. The move aimed to limit further exposure while the investigation continues. Fireworks claims the attackers were financially motivated, targeting tokens that grant access to large language models.

By selling these tokens to unauthorized third parties, the criminals can burn through AI model usage without using their own resources.

The company has not yet reported any evidence that the stolen tokens were used, but the potential for abuse remains high.

Our read is that this breach highlights the risk of relying on third‑party AI platforms.

Companies must audit data flows and enforce isolation to prevent a single point of failure from compromising their customers.

As more firms integrate external LLM services, the stakes for data security grow.

This is the second incident affecting Fireworks in as many months, following an October breach.

The pattern suggests attackers are honing in on LLM access points.

Stakeholders should demand transparency from providers and stay alert to new security advisories.