Law and data protectionAs of 11 October 2026René HerzerCreated with AI

Zero Data Retention: Is it enough for data protection?

Zero Data Retention (ZDR) means that the model provider does not store inputs and answers after processing. That lowers one risk, but on its own it is not enough. The data still leaves your organisation and is processed by the provider, the commitment often covers only certain features and requires approval, and third-country access, the data processing agreement and professional secrecy remain separate questions. ZDR is one part of your assessment, not a substitute for it.

What does Zero Data Retention mean?

Zero Data Retention (ZDR) is a commitment by a model provider: inputs and answers are not stored after processing, not even in logs for abuse detection. Without ZDR, many providers keep content for a period of time. At OpenAI, for example, according to its own documentation (as of October 2026), the default for the API is retention of up to 30 days for abuse detection; ZDR is only available there after prior approval by OpenAI and only for certain endpoints.

Which exceptions are usual?

Commitments differ from provider to provider. Typical exceptions are:

Only after approval: ZDR has to be requested and granted, often only for business customers.
Only for certain features: Saved conversations, uploaded files or search indexes at the provider remain stored until someone deletes them.
Legal obligations: Retention where a law or an order requires it.
Protection against abuse: Reservations allowing content to be kept after all if abuse is suspected.

For this reason, read the documentation for every endpoint and every feature your application uses, and have the approval confirmed in writing.

Why is ZDR not enough?

ZDR only governs what happens after the answer. The GDPR comes in earlier: the transfer and the processing at the provider are already processing within the meaning of Art. 4(2) GDPR.

Question	Does ZDR solve it?
Does the data leave your organisation?	No, it is processed by the provider.
Do you need a legal basis and a data processing agreement?	Yes, unchanged.
Is your data used for training?	A separate commitment, check it separately.
Is a US provider involved?	No. The US Cloud Act applies to US companies even in European data centres, and during processing the data is with the provider.
Is professional secrecy preserved?	No, processing without storage is still a disclosure under § 203 StGB.
Which features are covered?	Only those the provider names.
What does your own platform store?	Not covered: chats, documents and logs in the application need their own deletion rules.

The German Data Protection Conference (DSK, the joint body of the German data protection authorities) therefore distinguishes between closed systems, in which the data stays in a delimited environment, and open systems, in which inputs leave the protected area. It considers technically closed systems preferable.

What should you check instead?

Location: Where does the AI model run, in which data centre, operated by whom?
Role: Is the model provider your processor or sub-processor, with a data processing agreement?
Training: Contractually excluded?
Storage: For how long, for what purpose, who has access? This is where ZDR belongs.
Third country: Adequacy decision, standard contractual clauses, US Cloud Act (US Cloud Act and AI).
Protection of secrets: Agreement under § 203 StGB, for social data § 80 SGB X (German Social Code, Book X).
Your own platform: How long do chats and logs stay with you? (Logging and retention)

How does this work with basebox?

With the server, the AI model runs on your hardware; inputs go to no model provider, so the question of ZDR does not arise there. In the cloud, the AI model runs on our servers in the data centre in Munich (München-Ost); basebox is a German company, and the US Cloud Act does not apply to it. Your data does not train any model.

An external model is only available as a separate app with an external model, in a data centre in the EU. The provider does not use the inputs for training and logs them for abuse detection for up to 30 days. Because it is a US company, the US Cloud Act applies there. Your administrator shares such an app only with selected groups, and you decide together with your data protection officer for which data.

If basebox uses additional computing capacity from other providers, such as STACKIT, inputs are not stored there at all. Every such provider is listed among the sub-processors.

Not legal advice: Providers’ commitments change; the contract and the current documentation are authoritative. The assessment for your data is made by your organisation together with its data protection officer. Research as of 11 October 2026.

Frequently asked questions

Is Zero Data Retention a legal category?

No. It is a commitment by the provider, and each provider defines it differently. What counts is what the contract and the provider's documentation say: which features are covered, which exceptions apply and whether the commitment has to be approved.

Does a provider without ZDR store my data permanently?

Not necessarily. Many providers keep inputs for a limited time for abuse detection, often up to 30 days, and do not use them for training. What matters is whether the duration and the purpose are acceptable for your data.

Does ZDR help with § 203 StGB?

Only to a limited extent. For professional secrecy under § 203 StGB (the German Criminal Code), what counts is whether you disclose secrets and whether the service provider is bound to confidentiality. Processing without storage is still a disclosure. So you still need an agreement under § 203 StGB.

Sources

App with an external model Security and proof

This article was created with the help of AI, based on the sources listed. Responsible for the content: basebox GmbH. Not legal advice, no liability. Research as of 11 October 2026.