Notion has confirmed that a breach at its AI infrastructure provider, Fireworks AI, exposed chat transcripts from its Notion AI service, including messages that contained users' passwords and API keys. The company says fewer than 1% of its workspaces were affected, and it published an official security advisory on November 12, refreshed on November 13, with a day-by-day account of the exposure.

Notion's own systems were not breached: the data was exposed one layer down, at a provider most Notion customers had never heard of — one that had promised not to keep their data at all.

What the hackers took

Fireworks AI is a platform developers use to access open-weight AI models. Notion used it to serve some open-weight models in Notion AI and to run classification jobs for Insights, its AI-usage analytics feature. The stolen material was the Insights dataset: transcripts of Notion AI chats, containing any input to or output from the model that was written to the chat.

According to Notion's advisory, Fireworks alerted it on November 2 to an exposed dataset that first appeared to contain only synthetic test data. On November 3, Notion cut all connectivity with Fireworks and deleted its stored data. On November 6, Fireworks warned that customer data might have been accessed; on November 8, it confirmed the dataset held real Notion customer data. Notion began contacting affected customers on November 9.

Notion says that for the vast majority of affected workspaces, the transcripts were unattributable — not labelled with a user or workspace, and without tool-call results. But they still contained what users had typed. During its investigation, Notion found API keys and other credentials in the messages. On November 11, it revoked the exposed Notion API keys; on November 12, it reported other services' keys to their vendors. Affected customers should treat the data as exposed and rotate any passwords or credentials pasted into Notion AI.

The promise that did not hold

The most damaging detail may be what Notion says about the data itself: Fireworks had contractually committed to zero data retention for Notion-bound traffic — yet Notion's investigation found data was retained from the Fireworks endpoints used for Insights.

The attack began at a third-party cloud provider hosting Fireworks' research machines, which suffered its own security incident on October 23. On October 24, the attacker used that access to reach Fireworks development machines and run an unauthorised workload. Fireworks said the criminals took customer billing information and some datasets, and links the attack to a financially motivated group that monetises unauthorised access to large language models — a practice known as LLMjacking. The company has brought in Palo Alto Networks' Unit 42 to help with the investigation. It was the second intrusion at Fireworks in about a month: in early October, hackers accessed environment variables holding API keys and tokens for a limited number of accounts.

What Notion users should do now

Notion has disabled Insights and no longer serves open-weight models through Fireworks. Affected workspace owners can review the exposed data in Notion's web interface. The company recommends treating any password, API key, or other credential entered into Notion AI during the incident window as exposed, changing it, and checking login records for suspicious activity.

Fireworks serves tens of thousands of customers, including Harvey and Cursor. So far, Notion is the only customer confirmed affected — but the incident shows how a breach at a little-known AI infrastructure provider can quietly reach into a household-name product, and how a data-retention promise is only as strong as the systems behind it.