In May we wrote down what we do with your data, and the core of it was that model calls are pinned to zero-retention endpoints. That is still true, and it is still the default. What has changed is that the decision is now yours to make, per organisation, with the cost of it visible rather than buried.
The setting
Organisation settings > Default privacy mode. On means every model call your organisation makes runs in privacy mode: the provider does not train on your data and does not keep logs of it. Off means standard routing, which is cheaper, and where the provider may retain the request.
New organisations start with it on. Nobody had it quietly switched off underneath them.
It covers everything you are billed for
This was the harder half of the work. A privacy setting that only applies to one surface is worse than none, because it tells you that you are covered when you partly are not. The setting now applies to workflow runs, standalone agents, voice calls and Co-Worker alike, and the tier a run was made under is recorded with the run.
Two consequences follow from that. A call that fails over to a second model bills at the tier it started in, not at whatever the replacement happens to support. And if we cannot confirm a route meets the bar, the safe answer wins: it is treated as privacy mode.
Why show two prices
Privacy costs more. Zero-retention endpoints are priced above standard ones by the providers, and pretending otherwise would mean padding every price to hide it. Instead the model picker shows both numbers, so you can see what a model costs with privacy on and with it off before you choose one, and a credit range updates as you change your selection.
Most teams will leave it on and never think about it again. Some workloads genuinely do not need it, such as generating public marketing copy, and those teams should be able to take the cheaper route knowingly rather than discovering it in an invoice.
What has not changed
Your content is not used to train models in either mode. That commitment sits above this setting and is unaffected by it. Privacy mode is about what the provider is allowed to retain, not about who may learn from your data, and the answer to the second question is still nobody.
Every run still produces a full execution log, and the log now names the privacy tier alongside the model. As before, if your security review needs something more specific than this, write to us and we will answer it.