A Notion customer reports receiving an email about a Fireworks AI security incident, with no separate official Notion announcement found at the time. Notion said its investigation identified fewer than 1% of workspaces as potentially affected. The exposed material may include Notion AI chat transcripts used for the Insights feature. The transcripts had no labels identifying their authors or workspaces, and included no tool-call results; the transcript text itself was the only information that could reveal what users had discussed. Notion said no other workspace data was affected.
Notion used Fireworks for Insights classification and to provide open-weight AI models. After learning of potential incidents, it cut connections to Fireworks; on November 6, Fireworks warned that attackers might have accessed customer data, and on November 8 confirmed a dataset that Notion found contained the chat transcripts. Notion disabled Insights and stopped serving open-weight models through Fireworks while investigating. It advised affected customers to review the data available in their workspaces and rotate passwords, API keys, or other credentials that may have been pasted into chats. The reporting customer says they downloaded the dataset and found one of their chats in full, underscoring that “unattributable” did not mean the transcript content was absent.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.