Notion wasn't hacked. A company that ran its AI was, and some customers' chats went with it

Notion, the note-taking and planning app, published a security advisory on 10 November with an awkward message. Its own systems were never broken into, yet some customers' conversations with Notion AI ended up in a stranger's hands anyway.

The weak point was Fireworks AI, a company Notion paid to run some of the AI models behind its assistant. Fireworks was breached in October. On 2 November it told Notion that a small set of data had been accessed, and at first that looked like harmless test material. Notion cut the connection the next day and deleted what Fireworks held. Then, on 8 November, Fireworks confirmed the final list, and it included real chat transcripts kept for a feature called Insights, which shows workspace owners how their teams use AI. Notion puts the damage at fewer than 1% of workspaces, and says most of the transcripts carry no label showing who wrote them or which workspace they came from.

Affected customers were emailed on 9 November. "If you haven't heard from us, we haven't identified your workspace as affected, and no action is needed right now," the advisory says. For those who did get the email, the advice is blunt: treat the data as exposed, and if you ever typed a password, an API key or any other login detail into Notion AI, change it. Notion has since switched Insights off and stopped using Fireworks.

The lesson reaches well past Notion. A chat with an AI assistant feels private, but the words often travel on to other companies, so passwords don't belong in one. If you'd like to see where else your accounts and sign-ins might be loose, Tendvane's Privacy & accounts check is a quick place to start.