Home / News

Fireworks AI breach exposed Notion AI chat transcripts through a third-party vendor

Any business data you send to an AI tool passes through a chain of vendors, and the Fireworks AI breach shows that the weakest link in that chain sets your real level of security. If your contracts, invoices or client chats go through an AI service, find out which model provider sits behind it and what that provider stores.

Fireworks AI is a platform developers use to run open models. It confirmed that attackers broke into its cloud provider, reached a development system and took billing information and some customer datasets. According to TechCrunch, one affected customer is Notion. Chat transcripts from the Notion AI "Insights" feature were stolen, and about 1% of Notion customers were affected. Notion cut its connection to Fireworks on November 3. This is the second incident at Fireworks in two months.

What this means for a small business

Notion users never chose Fireworks. They chose Notion. That is the usual setup: a support bot, a lead assistant or an OCR service calls a model through someone else's infrastructure. When I build a pipeline on Claude or on open models, I map every hop the data takes. Where the task allows, I strip names and ID numbers before the model call. I also keep prompt logs on the client's server, not with the vendor. Fireworks says the attackers wanted to resell LLM access, so spending limits and a separate API key per workflow matter as much as encryption. I cover this in more detail in the data security part of my FAQ. One limit remains: none of this lets you audit a vendor's cloud, so you still depend on their security practices.

Why it matters: Ask every AI vendor who processes your data and how long they keep it. Then send the model only the fields the task actually needs.

Source: TechCrunch