Zero Data Retention AI
AI that processes your files to answer, then does not keep them.
Quick Answer
Zero data retention (ZDR) is an arrangement in which an AI provider processes your inputs and generates outputs without storing prompts, uploaded files, or responses once the request completes. It reduces exposure of sensitive project data, but its exact scope varies by vendor, so the contract terms need to be read carefully.
The Full Picture
When you send a drawing or a bid to a hosted AI model, the request passes through servers you do not control. By default, many services keep inputs and outputs for some period, for abuse monitoring, debugging, or product improvement. Zero data retention is an option or commitment under which that storage does not happen beyond what is needed to complete the request itself.
The term is not a single technical standard, and it is used differently across vendors. Some offer ZDR as an enterprise setting on specific endpoints, with exclusions for certain features. Others describe retention windows measured in days rather than zero. Questions to ask include: what exactly is not stored (prompts, files, outputs, metadata), whether logs are kept, which features are excluded, and whether the commitment is contractual or only a policy page.
ZDR also does not mean no risk. Data is still transmitted and processed in memory during the request, so transport encryption and vendor security still matter. The application you use may keep its own copy of your files and results separately from the underlying model provider. A tool can use a model with zero retention and still store your documents in its own database, which is a different policy to review.
For AEC teams, ZDR is one control within a wider data governance approach. It helps when handling client-confidential or security-sensitive documents, but it should sit beside access controls, deletion rights, and contract flow-down review. NIST guidance on media sanitization is a useful reference for what deletion should actually mean.
Real Examples
Common Misconceptions
People assume: Zero data retention means the AI never touches or sees my data.
Actually: The model must process your data to respond. ZDR is about what is stored afterward, not about whether the data is processed.
People assume: If the model has zero retention, the whole tool does too.
Actually: The application layer may store files, results, and logs independently. Review both the model provider's terms and the software vendor's own retention policy.
Frequently Asked Questions
What does zero data retention mean?
It means the AI provider does not store your prompts, uploaded content, or generated responses after processing the request, beyond what is needed to deliver the result.
Is zero data retention the same as no training on my data?
No. Not training on customer data is one commitment; not storing it at all is a stronger and separate one. Vendors may offer either, both, or neither, so check each in the terms.
Does ZDR make AI safe for confidential drawings?
It reduces one risk, but you still need encryption in transit, access controls, vendor security review, and attention to contract obligations to the owner.
How can I verify a vendor's ZDR claim?
Ask for the commitment in the contract or data processing terms, find out which features are excluded, and request security documentation. A marketing page alone is weak evidence.
Do all AI vendors offer zero data retention?
No. Availability varies by provider and plan, and it is often limited to specific enterprise arrangements or endpoints, so confirm it before relying on it.