Beepre reports that Lin Qiao, CEO of Fireworks, has published a detailed response addressing recent controversies over data retention. Previously, Fireworks was hacked, resulting in the leakage of chat logs from some Notion AI users. Notion stated that Fireworks had committed in its contract not to retain this data.
Jiao Lin explained that Fireworks' services are divided into two types:
The first type is standard AI inference. Users send a question, the model generates an answer and returns it, and the platform does not permanently store the input or output. Fireworks' Chat Completions, Nexus, and serverless training services all use this mode and support zero data retention (ZDR).
The second approach is batch processing. Customers first upload datasets, and then the platform runs tasks in batch. Since the platform does not know when customers no longer need these files, it retains them until the customer manually deletes them. Jiaolin emphasized that OpenAI, AWS, and Google also employ a similar design in their batch processing interfaces.
She drew an analogy: uploading a file to a cloud drive is different from entering numbers into a calculator. The former requires users to manage and delete files themselves, while the latter simply performs calculations and returns results.
Jiaolin stated that Fireworks will not retain copies after customers delete their data. For general inference and serverless training, the platform provides a zero-retention guarantee without requiring additional cleanup from customers.
The Fireworks official documentation also states that enterprise customers can enable a zero-data-retention policy. Once enabled, the platform will automatically reject batch tasks and file uploads that require data storage. However, this restriction is disabled by default and will not automatically delete previously stored files.
Jiao Lin believes that, whenever business allows, processing methods that do not require data storage should be prioritized. Every additional piece of data stored adds another burden that must be protected and cleaned up.