Fireworks AI Data Retention
6 steps1 pieces5 Sources
Fireworks AI, a US-based specialized intelligence platform, has clarified its data retention policies following a security breach that exposed Notion AI chat transcripts. The incident highlighted the distinction between two standard API designs: customer-managed retention (upload-use-delete) for batch jobs and ephemeral processing (send-compute-return) for real-time inference. CEO Lin Qiao stated on November 14, 2026, that Fireworks prioritizes ephemeral processing where workflows allow, as it eliminates human errors like forgotten datasets. However, the breach occurred in storage systems outside the default Zero Data Retention (ZDR) scope, specifically affecting batch inference and stored responses. While Fireworks offers an enterprise-wide ZDR policy to reject requests requiring persistence, this setting is off by default. The disagreement centers on whether contractual promises of non-retention cover all service modes or only specific endpoints, with Notion noting that data used for its Insights feature was retained despite expectations. Readers should monitor whether enterprise customers actively enable account-wide ZDR restrictions and how deletion processes for batch task data are verified.
How it got here
October 23, 2026: A third-party cloud provider hosting Fireworks R&D machines experienced a security incident.
venturebeat.com ↗October 24, 2026: Unauthorized parties accessed Fireworks development machines, leading to additional customer data impact.
venturebeat.com ↗