Notion AI Chat Transcripts Exposed in Fireworks AI Security Breach

Notion, the cloud-based workspace platform, discloses that a breach at its AI infrastructure provider, Fireworks AI, exposes chat transcripts from a small number of workspaces. Fireworks AI confirms that an attacker accesses its development environment on October 23 via a third-party cloud provider, compromising resource metadata, billing information, model weights, and datasets belonging to some customers. This marks the second breach for Fireworks in early October, when hackers previously access environment variables containing API keys and tokens.

Notion receives an alert from Fireworks on November 2 about an exposed dataset that supposedly contains only synthetic test data, prompting the company to cut off traffic to Fireworks, delete all vendor-hosted data, and halt training runs by November 3. However, on November 8, Fireworks confirms that real customer data was accessed. Notion's analysis of the dataset reveals it includes Notion AI chat transcripts from the Insights feature, which gives workspace administrators visibility into how their teams use AI. That feature is now disabled.

Notion states that for the vast majority of impacted workspaces, the transcripts are unattributable, meaning they lack labels identifying who wrote them or which workspace they came from, and they do not contain tool call results. The company notes that Fireworks had contractually committed to zero data retention for Notion data. Fewer than 1% of Notion users are affected, and workspace owners can review potentially impacted data through the Security incident review tool in Notion's desktop web interface.

Read More at the original source →