Notion says Fireworks AI breach exposed Notion AI chat transcripts despite a zero-retention promise

Notion says the security incident at its AI vendor Fireworks AI exposed Notion AI chat transcripts from fewer than 1% of Notion workspaces, according to a disclosure on its help center. Any input to or output from the model that was written to an affected chat is included, and Notion tells customers to treat the data as exposed and to rotate any passwords or API keys they pasted into Notion AI.

Notion used Fireworks to run the analysis behind Insights, a feature that shows administrators how their workspace uses AI, and to serve open-weight models in its product. Fireworks had given Notion a contractual commitment not to retain data, but Notion's investigation found that data was kept from the Fireworks endpoints it used. On 3 November Notion disconnected from Fireworks, had all its data there deleted, disabled Insights and halted training runs. It is now auditing critical vendors such as inference providers on data retention.

Notion is a confirmed customer victim of Fireworks' second breach, in which an attacker went through Fireworks' cloud provider to reach a development system and took billing information and some customer datasets, TechCrunch reported. Fireworks says its customers include Harvey and Cursor.