[Fireworks CEO Responds to Data Retention Controversy: Regular Inference Doesn't Store Data, Batch Tasks Require Customers to Delete Them Themselves] Beating AI News reports that Fireworks CEO Lin Qiao has published a lengthy response to the recent data retention controversy. Previously, Fireworks was hacked, and chat logs of some Notion AI users were leaked. Notion pointed out that Fireworks had promised in its contracts not to retain this data. Qiao explained that Fireworks' services are divided into two types: The first is regular AI inference. Users send questions, the model generates answers and returns them; the platform does not store input and output for long periods. Fireworks' Chat Completions, Nexus, and serverless training services all use this model, supporting zero data retention (ZDR). The second is batch processing. Customers upload datasets first, and then the platform runs tasks in batches. Because the platform doesn't know when customers will no longer need these files, it will keep them until the customers actively delete them. Qiao emphasized that OpenAI, AWS, and Google's batch processing interfaces also use similar designs. She used an analogy: uploading a file to a cloud drive and entering numbers into a calculator are two different things. The former requires users to manage and delete files themselves, while the latter only requires calculation and returning the result. Qiao Lin stated that Fireworks will not retain copies of data after a client deletes it. For regular inference and serverless training, the platform provides a zero-retention guarantee, requiring no additional cleanup from the client. Fireworks official documentation also indicates that enterprise clients can enable a zero-retention limit for all accounts. Once enabled, the platform will directly reject batch tasks and file uploads that require data retention. However, this limit is off by default and will not automatically delete previously saved files. Qiao Lin believes that as long as business allows, processing methods that do not require data retention should be prioritized. Every additional copy of data retained adds to the burden of protection and cleanup. [Original Link]