Fireworks AI Hit by Second Cyberattack, Exposing Notion Chat Data
Fireworks AI suffered another breach, exposing Notion AI chat transcripts in weeks.
Why it matters: Legal professionals rely on AI tools like Fireworks and Notion to manage sensitive client data securely. Repeated breaches highlight risks in third-party AI providers' data handling and prompt urgent scrutiny of cybersecurity practices.
- Fireworks AI had a security breach on October 6, 2026, with unauthorized access to internal credentials.
- A second breach on October 23, 2026, via a third-party cloud provider exposed Notion AI chat transcripts stored on Fireworks' R&D machines.
- Less than 1% of Notion workspaces were affected, with chat data related to the Insights feature.
- Fireworks typically follows a zero data retention policy, but stored responses and batch jobs allowed persistence of some customer content.
Fireworks AI experienced two separate cyberattacks in October 2026, compromising sensitive customer data including Notion AI chat transcripts used for the Insights feature. The first security incident on October 6 involved unauthorized access to internal credentials and affected a limited number of Fireworks customer accounts, according to the company's security bulletin.
The second breach occurred on October 23, when a third-party cloud provider hosting Fireworks AI's research and development machines was compromised. This allowed attackers to access customer data, including stored Notion AI chat transcripts. Notion confirmed in its own security advisory that fewer than 1% of its workspaces were impacted. The exposed data mainly consisted of chat logs used for the Insights feature, which Notion described as largely unattributable to specific users.
Fireworks AI asserts it has a default zero data retention policy to protect customers, but features like batch jobs and stored responses require storing some customer content. This stored data was accessible in the October 23 breach and led to exposure of Notion chat transcripts. Industry analysis by VentureBeat outlined how these exceptions to data retention policies can introduce vulnerabilities.
The attackers behind the October 6 breach were linked to a cybercriminal group engaging in 'LLMjacking'—selling unauthorized access to large language models—though detailed attribution was not disclosed. By November 9, Notion had contacted affected customers to inform them of the breach.
These incidents highlight critical concerns within the legal sector, where AI-powered tools increasingly handle confidential client communications. Continuous data security lapses by third-party AI vendors like Fireworks AI call for heightened due diligence by legal professionals to safeguard sensitive data.
By the numbers:
- October 6, 2026 — First Fireworks AI breach date
- October 23, 2026 — Second breach via third-party cloud provider
- Less than 1% — Proportion of Notion workspaces affected by the breach
Yes, but: While fewer than 1% of Notion workspaces were affected, the breach exposed chat data that should have been protected under Fireworks' zero data retention policy due to stored responses and batch jobs.
What's next: Fireworks AI and Notion have yet to disclose specific measures planned to prevent future breaches, leaving ongoing cybersecurity improvements to watch.