Fireworks AI breach exposes Notion customer chat transcripts, marking second incident in two months
A cyberattack on Fireworks AI's cloud infrastructure leaked billing data and chat transcripts from Notion customers, affecting roughly 1% of Notion's user base.
Last verified:
Fireworks AI, a platform developers use to access open-source AI models, disclosed a security breach on its cloud infrastructure that exposed customer billing data and datasets. According to TechCrunch, the attack also compromised chat transcripts belonging to Notion customers who used Notion’s AI Insights feature, affecting roughly 1% of Notion’s overall user base. The incident underscores how security vulnerabilities at infrastructure vendors can cascade across their customer base.
Fireworks AI Breach Details
According to TechCrunch, an attacker breached Fireworks AI’s cloud provider infrastructure to gain access to a Fireworks development system. Fireworks AI serves tens of thousands of customers worldwide, including legal tech company Harvey and code editor Cursor.
The compromised data included customer billing information and datasets. For Notion specifically, TechCrunch reports that the breach exposed chat transcripts generated by users of Notion’s AI Insights feature. Notion confirmed in a disclosure published this week that the stolen transcripts include model inputs and outputs written to the chat, though the company noted that much of the data was unattributable and could not be traced to specific customers.
On November 3, Notion disconnected its systems from Fireworks, deleted the data stored in its systems, and terminated all connectivity between the two platforms, according to TechCrunch.
Financial Motive and Token Theft
According to TechCrunch, Fireworks AI believes the attackers are financially motivated and focused on stealing access tokens for large language models (LLMs). The goal, Fireworks says, is to sell these tokens to unauthorized third parties who can then consume AI model usage without depleting their own API budgets.
This theft pattern is known as token-burning schemes, where compromised API credentials are monetized through resale or unauthorized consumption.
Second Incident in Rapid Succession
According to TechCrunch, Fireworks reports this appears to be its second security incident in two months, following a breach in October 2026. The source material does not provide details on the scope or nature of the October incident.
Why This Matters
This breach highlights a concrete trade-off for Notion customers: users of Notion’s AI Insights feature now face uncertainty about whether to continue using the feature pending Notion’s confirmation of an alternative vendor with verified security posture. Organizations that rely on Notion for internal documentation workflows must decide whether the 1% exposure risk documented by Notion justifies disabling AI-powered chat transcripts until new assurances are in place.
More broadly, the incident reveals how third-party dependency in AI infrastructure creates supply-chain vulnerabilities. When a single vendor (Fireworks AI) serves as an intermediary between enterprises and LLM APIs, a breach at that intermediary affects not just Fireworks’ direct customers but all downstream end-users of those customers’ products. Enterprises selecting AI service providers must now weigh feature richness against vendor security incident history and breach disclosure timelines.
Frequently Asked Questions
What data did the Fireworks AI breach expose?
According to TechCrunch, hackers accessed Fireworks customer billing information and datasets. For Notion specifically, the breach exposed chat transcripts from users of Notion's AI Insights feature, affecting approximately 1% of Notion's customer base.
How did the attackers gain access to Fireworks AI?
TechCrunch reports that the attacker breached Fireworks AI's cloud provider infrastructure to gain unauthorized access to a Fireworks development system, which then allowed access to customer data.
What is the suspected motive behind this breach?
Fireworks AI believes the attackers are financially motivated, with goals focused on selling tokens from compromised large language model access to unauthorized third parties.
Is this the first breach at Fireworks AI?
No. According to TechCrunch, Fireworks reports this appears to be its second security incident in two months, following a breach in October 2026.