Considering their core service is inputs/outputs and this is a common question from developers, for your own sake, you should assume that there’s some degree of logging and access.
This doesn’t mean that it is actually happening, but it’s common for companies to be vague on points that they expect users won’t like. Other services are explicit because they know it will enhance their marketability to developers.
Since LLMs are one their specialties, it’s very unlikely this point wasn’t considered, and it doesn’t take two months to get a response from a legal team. Responding “we don’t log inputs and outputs” wouldn’t require a legal team if they actually don’t store/log the inputs and outputs.
So without additional information, I would assume that it is stored for some unknown period of time and is NOT safe to use with confidential information.