Availability. The Pre assistant (“Pre AI”) and the on-device intelligence utilities described in this section are not included in the current version of the pre app. Nothing in this section happens unless your app version offers these features; when we make them available, we will update this policy first and the rules below will apply.
Some optional intelligence utilities can run locally on supported devices in app versions that include them. These may include automatic alt text, safety pre-checks before you post, and flyer scan, which turns a photo of an event flyer into an event draft. The local processing step does not send the content it reads to pre or OpenAI. If you then choose to post, send, or save the result, the content you submit reaches our servers like other content and the rest of this policy applies, including our server-side safety checks.
You choose how Pre AI runs. Cloud Pre, which is included with Host Pass, sends bounded text to OpenAI. On supported Apple devices, you can instead choose the more private on-device model, or turn Pre AI off. Changing from Cloud to on-device or off also withdraws your prior Cloud Pre permissions for private conversations.
When you invoke Cloud Pre or keep it active in a direct or group conversation, pre sends OpenAI a bounded amount of text needed to answer: participant display names, your request, and limited recent text context from that conversation. OpenAI also receives a pseudonymous safety identifier derived from the requesting account; this is not the account’s raw pre user ID. Every current participant must first agree for that conversation. The feature does not send the contents of photos, videos, audio, or other media attachments to OpenAI, and it does not send messages from unrelated conversations or former participants. Pre’s reply is stored in the conversation like other chat content so participants can see it. Anyone can revoke permission, and you can pause Pre in that conversation.
Requests to OpenAI use store: false, which means the response is not retained as OpenAI application state. OpenAI says API data is not used to train or improve its models unless the API customer opts in. Under OpenAI’s standard data controls, however, bounded prompt and response text may still be included in abuse-monitoring logs for up to 30 days, or longer when required by law or reasonably necessary to protect people or services. This is separate from pre’s own retention. You can read OpenAI’s API data controls.