Glossary
Zero Data Retention (ZDR) — Definition for Lawyers
Zero data retention means the AI vendor stores nothing after replying. Which plans offer ZDR, the Covered Models exception, and why it matters for privilege and GDPR.
On this page
Zero data retention (ZDR) is a contractual arrangement under which an AI provider does not store your prompts, files or the model's outputs once the response has been delivered: nothing is logged for abuse monitoring and nothing sits on the vendor's servers to be subpoenaed. It is stronger than "we do not train on your data", and as of September 2026 it exists only on API and enterprise contracts.
No training is not no retention
The two promises get conflated in vendor decks and in firm AI policies, and the difference is the whole point. "No training" means your material will not be used to improve the model; it says nothing about how long the vendor keeps it. OpenAI's Business, Enterprise, Edu and API tiers do not train by default, but the API still retains content for 30 days for abuse monitoring; ZDR is a separate contractual term for eligible customers. Anthropic's API does not retain conversation content by default and says retained data is never used for training without express permission, but ZDR is arranged per organisation and flagged content may be kept for up to two years. During any retention window the material can be read by the vendor's trust-and-safety staff, exposed in a breach, or produced under legal process.
| Provider and plan | Trains by default? | Standard retention | ZDR available? |
|---|---|---|---|
| ChatGPT Free, Go, Plus, Pro | Yes, unless the user opts out | Consumer terms apply | No |
| ChatGPT Business, Enterprise, Edu; OpenAI API | No | API: 30 days for abuse monitoring | Contractual for eligible customers |
| Claude Free, Pro, Max | Only if the user opted in | 5 years opted in; 30 days opted out | No |
| Claude Team and Enterprise chat | No | Flagged content up to 2 years | No; the chat interface is not ZDR-eligible |
| Claude API; Claude Code under Enterprise | No | Not retained by default | Yes, per organisation, except Covered Models |
| Claude Fable 5.1 and Mythos 5.1 (Covered Models) | No | 30-day retention required | Not without Anthropic's express authorisation |
| Meta Muse Spark API | Contributor tier retains data; standard tier does not | Tier-dependent | Not offered as ZDR in the sources reviewed |
| Consumer Gemini Apps | Yes, unless Keep Activity is off | 18 months by default; a subset read by human reviewers | No |
OpenAI's Astra for Law, announced 17 September 2026, ships with a zero-data-retention API alongside its 26 partner plugins. The counter-intuitive entry in the table is Anthropic's: Claude Fable 5.1, first on the Vals Index of 15 September 2026, is a Covered Model that requires 30-day retention, so a firm that needs ZDR for a matter is choosing Opus 5 or Sonnet 5 on the API. The confidentiality checklist has the plan-by-plan detail for every major tool.
Why retention decides privilege and discovery
Two rulings from the Manhattan federal court in the same winter show what a retention window costs. In United States v. Heppner (S.D.N.Y. 2026), with a bench ruling on 10 February and a written opinion on 17 February, Judge Rakoff held that a defendant's research memos produced with consumer Claude were neither privileged nor work product: "Claude is not an attorney", and Anthropic's consumer terms permitted logging and disclosure, so privilege never attached in the first place. In In re OpenAI Copyright Litigation (S.D.N.Y.), Magistrate Judge Wang ordered production of 20 million de-identified ChatGPT logs in November 2025 and Judge Stein affirmed on 5 January 2026, reasoning that users had voluntarily submitted their communications to OpenAI. Logs the vendor holds are discoverable ESI. Logs the vendor never kept cannot be. The UK Upper Tribunal made the same point from the other side of the Atlantic in UK v SSHD [2026] UKUT 81 (IAC): pasting client letters into public AI tools puts them "in the public domain". The privilege and AI page traces the case law.
For European firms ZDR is also a GDPR answer. Storage limitation is easier to demonstrate when the processor stores nothing, and the Austrian bar's rule that client data stays on EU servers and Germany's BRAK position that tools storing client data for training breach § 43a BRAO both become simpler to satisfy. The GDPR and AI tools page covers the DPIA.
What ZDR does not cover
ZDR governs the vendor's copy. It does not delete the transcript on your own machine, the copy in your document management system, or the audit feed your compliance team switched on: Anthropic's Compliance API activity feed and local session transcripts from Cowork and Claude Code are retained for six years by default. It does not govern a subprocessor you did not know about; Microsoft's Legal Agent for Word requires Anthropic to be enabled as a subprocessor. And a ZDR API is not what your associates use at eleven at night; the consumer app on their phone is, which is why the privacy settings walkthrough exists.
How to ask for it
Put these six questions to the vendor in writing and keep the answers with the contract:
- Is zero data retention available on the plan quoted, and is it a signed contract term or a dashboard setting?
- Which models are excluded from ZDR, and what is their retention period?
- What is retained anyway (flagged content, abuse monitoring, telemetry), for how long, and who can read it?
- Which subprocessors receive prompts or outputs, and are they bound by the same term?
- Where are the servers, and can the vendor commit to an EU region?
- What happens to retained data on termination, and will the vendor certify deletion?
The vendor due-diligence page has the longer questionnaire. Module 6 of the AI Lab for Lawyers course walks through the retention terms of each major tool and where the settings live.
Frequently asked questions
What does zero data retention mean for AI tools?
It means the provider deletes your prompts, uploaded files and the model's outputs as soon as the response has been delivered, keeping no copy for abuse monitoring, support or training. It is a contractual commitment, not a dashboard setting, and as of September 2026 it is offered on OpenAI's API and enterprise contracts for eligible customers and arranged per organisation on Anthropic's API. Consumer plans, and the Claude Team and Enterprise chat interfaces, do not offer it.
Is no training on my data the same as zero data retention?
No. No training means your material will not be used to improve the model; the vendor may still keep it for weeks. OpenAI's API retains content for 30 days for abuse monitoring even though it does not train on it, and Anthropic may keep flagged content for up to two years. Zero data retention removes the copy altogether. For privileged litigation material and regulated data, the retention window is the risk, not the training.
Which AI models cannot be used under zero data retention?
Anthropic's Covered Models, meaning Claude Fable 5 and 5.1 and Mythos 5 and 5.1, require 30-day retention and are not ZDR-eligible unless Anthropic expressly authorises it. A firm that needs ZDR for a matter is, as of September 2026, choosing Claude Opus 5 or Sonnet 5 on the API, or an OpenAI enterprise or ZDR API arrangement such as Astra for Law. Check the exclusions before assuming a ZDR contract covers the model your lawyers actually use.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Guides
Which AI Tools Train on Your Data? A Confidentiality Checklist for Lawyers
Which AI tools train on your data, plan by plan as of September 2026: ChatGPT, Claude, Gemini, Copilot, legal vendors, plus bar rules and a 10-point checklist.
11 min read
Regulation & Ethics
Attorney–Client Privilege and AI: Heppner, Warner, the NYT Logs and Public Chatbots
Consumer chatbot memos are not privileged. US v. Heppner, Warner v. Gilbarco and the 20-million-log OpenAI order, plus rules that keep privilege intact.
10 min read
Regulation & Ethics
GDPR and AI Tools for Lawyers: Lawful Basis, DPIAs and Vendor Terms
How GDPR applies when law firms use Claude, ChatGPT or Copilot: controller and processor roles, lawful basis, DPIA triggers, transfers and vendor terms.
11 min read
Regulation & Ethics
AI Vendor Due Diligence for Law Firms: The Questions to Ask Before You Sign
AI vendor due diligence for law firms starts with the subprocessor chain, then training and retention terms, ZDR eligibility, data residency, audit logs and exit rights.
11 min read
Questions
Does Using ChatGPT Waive Attorney–Client Privilege?
Using ChatGPT rarely waives privilege in the technical sense; the courts' answer is worse. In Heppner, consumer-chatbot memos were never privileged at all.
6 min read
Glossary
Open-Weight Model — Definition for Lawyers
An open-weight model publishes its trained weights for anyone to run. Open weights vs open source, DeepSeek, Mistral and Muse Glimmer licences, self-hosting and government bans.
5 min read