Notion AI Customer Chat Data Compromised in Fireworks AI Hack

Notion, the popular collaborative workspace and productivity platform, is notifying a limited number of customers that their Notion AI chat transcripts were compromised in a data breach affecting its third-party AI inference provider, Fireworks AI.

The incident stems from two separate hacks against Fireworks AI in October. The second intrusion, occurring on 24 October, involved attackers executing an unauthorised workload that led to the compromise of customer data hosted on Fireworks AI’s R&D machines through a third-party cloud provider. Fireworks AI contained the incident on the same day and notified Notion on 2 November.

Prior to the breach, Notion utilised Fireworks AI as a general inference provider for its open-weight AI models, specifically impacting data related to the Notion AI Insights feature, which assists customers in understanding their AI usage.

Fireworks AI Compromise Affects Notion Data

Notion’s internal investigation confirmed that the hackers gained access to Notion AI chat transcripts. However, the company stated that fewer than 1% of Notion workspaces were affected. A significant portion of these compromised transcripts were unattributable to specific customers or their workspaces.

“Beyond what’s in the chat transcript itself, there’s no information tying the chat back to workspaces, users or companies,” Notion said in its customer notification. The company has contacted affected customers with guidance on securely retrieving and reviewing data within their workspaces, though it did not disclose the precise number of impacted customers.

Notion has advised affected users to treat the exposed data as compromised. “If you pasted sensitive information, passwords, API keys, or other credentials into Notion AI during the incident, rotate any exposed credentials,” the company cautioned its customers.

The second incident at Fireworks AI has been linked to a recent rise in what is being termed ‘LLMJacking’ activity, a category of cyberattack targeting large language models (LLMs) and their underlying infrastructure.

The initial Fireworks AI compromise in October occurred via exploited internal credentials, with the follow-up incident then leading to the Notion AI data exposure.