How AI is used on support conversations
When you message a company that uses Runner for support, the text of that conversation may be read by an AI model to work out what it is about and how urgent it is. This page says which provider that is, what reaches them, how long they keep it, and what happens if you ask for your data to be deleted.
This page describes what the software does. It does not tell you what any law requires of the company you were messaging.
What happens to your conversation
- Who receives it
- Anthropic
- What it is used for
- Reading a support conversation and suggesting how it should be categorised and prioritised. Nothing is sent to the customer as a result without a person approving it.
- What is sent
- The messages in that one conversation, and nothing else.
- What is not sent
- Contact details we hold as separate fields — email address, phone number, name, job title, country — are not included in what is sent. The message text itself is sent as written, so anything a person typed into a message goes with it. We do not currently strip personal details out of message text before it is sent.
- How long they keep it
- Anthropic deletes the conversation text from its systems within 30 days of receiving it.
- Deletion requests
- Anthropic publishes no way to delete an individual message on request. The copy is removed when its retention window ends.
- Training
- Anthropic's commercial terms state it does not train its models on customer content sent through the API.
If you ask for your data to be deleted
Anthropic does not offer a way to delete an individual conversation from its systems on request. When someone asks us to delete their data, we delete our own copy straight away and record the date Anthropic's copy runs out, so we can tell them exactly when it will be gone rather than claiming it already is.
Ask the company you were messaging — they hold the relationship with you and can act on it. When they do, the record of what was sent to the provider is settled at the same time, so the answer they give you about the provider’s copy comes from a recorded date rather than an estimate.
When the provider keeps it for longer
The provider documents two situations where content is kept beyond the ordinary window. Neither is something we choose or control, and both are worth knowing about:
- Content flagged under Anthropic's usage policy may be kept for up to two years.
- Trust-and-safety classification scores derived from content may be kept for up to seven years.
The full provider list
Every outside provider that processes customer conversation content, and the terms their handling of it sits under, is listed on the providers page.