Fireworks AI suffers second data breach, exposing Notion customer data
Fireworks AI has confirmed a second security breach in less than three weeks, exposing customer billing information and some datasets.
Fireworks AI runs a platform that developers use to access and run open AI models. According to the company's latest security advisory, the attacker did not break into Fireworks directly. Instead, they compromised a third-party cloud provider that hosted Fireworks' research and development machines.
How the attackers got in
That foothold, used on October 24, gave the intruder access to a Fireworks development system. From there, some customer data was within reach. The exposed information includes billing details such as billing addresses, invoice history, and account balances. Resource metadata was also taken.
For a limited number of accounts, the damage went further. Datasets, secrets, and incomplete model weights were also exposed.
October 6 incident
The October 24 breach followed an earlier incident detected on October 6. That one involved unauthorized use of Fireworks' internal credentials and secrets. Fireworks said the first breach did not expose customer models. The company initially downplayed how severe the access was.
Notion's response
Notion has been the most public about its exposure. The productivity company said less than 1% of its workspaces were affected. The exposed data primarily consisted of AI chat transcripts from Notion's Insights feature. Notion described those transcripts as unattributable, meaning they were not tied to identifiable users.
On November 3, it severed ties with Fireworks, deleted all data stored with the vendor, and disabled the Insights feature entirely.
AI supply chain vulnerability
Fireworks says it serves tens of thousands of customers worldwide. Many end users of Notion likely had no idea a company called Fireworks was processing their AI chats in the background. In this case, the weak point sat two steps removed from Notion's own systems.
Customer impact
Cursor and Harvey are both named as affected customers. The exposure of secrets for some accounts adds another layer of cleanup. Secrets typically include API keys and access tokens, so affected customers may need to audit what those credentials could reach.