A chat transcript can lose its author label and still contain a password. Notion said data exposed in a breach at Fireworks AI included Notion AI transcripts used for Insights, a feature that helped customers understand how they used AI. Notion says fewer than 1% of its workspaces were affected. Notion’s advisory
Fireworks handled the classification work behind Insights and also served open-weight models for Notion. Notion said it disconnected from Fireworks on Nov. 3. Fireworks confirmed the affected dataset on Nov. 8 and Notion began contacting affected customers on Nov. 9. The advisory says the transcripts lacked labels identifying the author or workspace for the vast majority of affected workspaces. It also says the transcripts excluded results returned by connected tools. A person could still have typed identifying details or credentials into a message. Notion’s timeline · TechCrunch’s reporting
Notion disabled Insights during the investigation. It says it revoked exposed Notion API keys it found in affected user messages on Nov. 11. Customers it notified can follow the email instructions to retrieve the affected data securely inside the workspace. Notion advises those customers to rotate sensitive credentials they pasted into Notion AI during the affected period. Notion’s response and instructions
In an affected workspace, the person who can review the transcript and the person who can replace a credential may be on different teams. Have the workspace owner tell each service owner which credential appeared in a transcript and where the service uses it.