What is zero data retention?

When software sends a customer's message to an AI model, a copy of that message travels to the model provider. Zero data retention is a promise about what happens to that copy afterward: nothing is kept. The term comes up in security reviews, vendor questionnaires and AI training courses, and it is often misunderstood as a promise about every system involved.

What zero data retention means

Zero data retention (ZDR) means a provider uses the data in a request only to produce the response, then stores none of it. The prompt, the response and any related content are not written to logs, not kept for later review and not used to train models.

The usual quiz answer to "what does zero data retention mean?" is the one that says the prompt, its responses and all related data are forgotten once the answer is returned. It does not mean a person deletes the data by hand afterward, and it does not mean the data is masked while it is stored. Masking and redaction are separate techniques; ZDR is about not storing at all.

No data retention and zero retention mean the same thing. A zero data retention policy is the written commitment, usually in a contract or data processing agreement, that spells out which services and endpoints it covers.

How zero data retention works with AI models

Without ZDR, an AI model provider may keep requests for a period after answering, commonly to monitor for abuse, debug problems or, under some consumer terms, improve its models. Under ZDR the flow is shorter.

1. The application sends the request: instructions, the customer's message and any context.
2. The model provider processes it in memory and returns a response.
3. The provider discards the request and response instead of logging or storing them.

ZDR is usually something a business customer requests and the provider approves, and it often applies only to certain models or endpoints. Features that need storage by design, such as saved conversation state, uploaded files or fine-tuning data, are typically outside it. That is why the exact scope matters more than the label.

What zero data retention does not cover

ZDR covers one provider and one hop. An AI customer service product usually involves several places data can live:

• The AI model provider, where ZDR typically applies.
• The vendor's own application, which stores conversations so your team can read and answer them.
• Search indexes built from your help articles and, sometimes, past conversations.
• Application logs, analytics and error monitoring.
• Your own helpdesk, CRM and email.

The practice this page rejects is treating "zero data retention" as a claim that nothing is stored anywhere. A customer service platform has to keep conversations to be a helpdesk at all. The honest version names which provider runs at ZDR and says plainly what the vendor itself keeps, for how long and how it is deleted.

In Aide, the conversations are kept because the team works from them: each one shows its Action Trace, with the sources under every reply, order actions and handoff notes, and the audit log records every change to the workspace with who made it, when and a before and after diff, kept for two years.

Zero data retention and AI agent security

ZDR reduces one risk: a copy of customer data sitting with a model provider. It does nothing about what an AI agent is allowed to do. Agent security also depends on who can change the agent's instructions, which actions it can take without a person, and whether every action can be traced afterward.

With Aide, roles decide who can change settings (Owner, Admin, Agent), refunds wait for the team by default, and the audit log shows every change to agents, procedures, knowledge and refund approvals, including changes made through API keys.

Questions to ask an AI customer service vendor

Whether you are comparing AI support tools or reviewing one you already use, ask:

1. Which AI model providers process our customer conversations, and does ZDR apply to each of them?
2. Does ZDR cover every feature we will use, or only some models or endpoints?
3. Is it written into the data processing agreement, or only stated on a web page?
4. What does your own platform store: conversations, files, search indexes, logs? For how long?
5. Is our data used to train any shared model, yours or a provider's?
6. How do you handle a customer's request to see or delete their data?
7. Who on your team and ours can see conversations, and is access logged?

In Aide, customer data requests and erasure requests from Shopify are handled automatically: a data request emails the workspace owners an export, and an erasure deletes that customer's conversations, messages, files and orders. See GDPR and customer data in Aide.

Where Aide fits

Aide, the AI platform for customer service, keeps the records a support team needs and makes them inspectable. Every conversation carries its Action Trace, the audit log records every change with a before and after diff, and customer data and erasure requests from Shopify are handled without anyone doing it by hand. Roles limit who can change the agent. On the model side, Aide's Claude models run at zero data retention through Anthropic's API, so the conversations they process are not stored by Anthropic or used to train its models, and open-weight models run on Aide's own servers. Aide is SOC 2 Type II certified and GDPR and HIPAA compliant; the security page covers hosting and the data processing agreement.