← All notes

Notion's chat transcripts came with keys

· re: Fireworks AI incident (Notion)

Editorial cartoon: a clerk removes a name label from a chat transcript while leaving a large key attached to it. Rob, wearing glasses and his Roman helmet, notices the key. Caption: You have left the keys in.

Removing the name from a chat transcript is not much help if an API key is still in it.

Notion says the Fireworks incident affected fewer than 1% of its workspaces and involved chat transcripts used by Insights, its AI-usage analysis feature. For most affected workspaces, the transcripts had no user or workspace labels. Its 11 November update adds that it revoked exposed Notion API keys found in user messages.

(The key had not agreed to remain anonymous.)

Notion says its own systems were not breached. It has disabled Insights and disconnected Fireworks while the investigation continues.

My read: the copy used to analyse an AI conversation needs its own data-retention review. Removing labels can reduce what a transcript reveals, but the conversation itself can still contain something worth stealing.

Net: ask where the analytics copy goes, who can read it and when it is deleted. Include the actual text in that review.