AI Agents for Lawyers

Questions

Is ChatGPT GDPR-Compliant for Law Firms? A Practical Answer

ChatGPT can be used GDPR-compliantly by a law firm on Business, Enterprise or API terms with training off, a processor contract and a DPIA; consumer plans cannot. Seven steps.

By Dr. Niklas SchmidtUpdated 13 November 20267 min read
On this page
  1. The yes/no matrix by plan, as of September 2026
  2. What the Garante fine and its annulment do and do not tell you
  3. Your own processing: where the firm is the controller
  4. Professional secrecy is stricter than the GDPR
  5. The seven-step compliance route
  6. Where this leaves you

ChatGPT is not GDPR-compliant or non-compliant in itself; your deployment is. On the consumer plans, which train on conversations unless the user opts out and come without a processor contract, client data cannot lawfully go in. On Business, Enterprise or API terms, with training off, retention set, a signed Art. 28 agreement and a DPIA, a law firm can use it. Professional secrecy rules add a stricter layer.

The yes/no matrix by plan, as of September 2026

PlanTrains on your inputs?RetentionContractClient data?
Free, Plus, Pro (Go is a regional consumer tier)Yes, unless the user opts out (retention overview)User-side deletion only; provider copies persist, and tens of billions of logs were preserved in the OpenAI copyright litigationConsumer terms; no Art. 28 processor contractNo
Business ($25 per seat monthly, $20 annual; Premium $125/$100, pricing as of 9 September 2026)No, by defaultWorkspace-configurableBusiness terms; verify the current processor paperwork on OpenAI's enterprise privacy pageYes, after the seven steps
Enterprise (custom)No, by default; GPT-6 Astra is off until admins enable it; ChatGPT Work carries Compliance API audit trailsAdmin-setEnterprise agreementYes, after the seven steps
API (your own tools)No, by default30 days for abuse monitoring; zero data retention is contractualAPI terms plus processor agreementYes; prefer ZDR where eligible
Astra for Law (announced 17 September 2026)Zero-data-retention API and "Trusted Access", per OpenAIVendor-statedNot yet independently evaluatedToo new to say

Two things the matrix cannot show. First, "no training by default" is a promise about model training, not about retention, human review for abuse monitoring or sub-processors; those are separate lines in the contract. Second, the plan tells you what OpenAI does; the GDPR asks what you do. The ChatGPT review covers the features to switch off; the rest of this page is about the firm's own obligations.

What the Garante fine and its annulment do and do not tell you

The Italian Garante fined OpenAI €15 million (decision 2 November 2024, announced 20 December 2024) and ordered a six-month radio and television information campaign. The Court of Rome conditionally suspended the fine in March 2025, as Lewis Silkin reported, and on 18 March 2026 annulled it entirely, with reasons published on 28 May 2026. The ground, as ppc.land summarised, was competence: the Irish Data Protection Commission became OpenAI's lead authority on 15 February 2024, before the Garante decided, so the one-stop-shop stripped the Garante of jurisdiction. The European Law Blog calls the result an "establish, then escape" enforcement gap.

For a law firm the annulment is neither reassurance nor alarm. Nothing was decided about lawful basis for training, transparency or age verification; those questions now sit with Dublin. What matters more is the EDPB's Opinion 28/2024 of 17 December 2024, which says an LLM is anonymous only if extracting personal data from it is "insignificant", sets a three-step legitimate-interest test, and expects deployers of a model trained unlawfully to assess that upstream unlawfulness. The Hamburg DPA's discussion paper of 15 July 2024 takes the friendlier view that LLMs do not "store" personal data, but adds the part that lands on the firm: data-subject rights bite on inputs and outputs at deployer level. Either way, the firm answers for what it typed in and what came out.

Your own processing: where the firm is the controller

On a business tier the firm is the controller and OpenAI the processor for the content of prompts and outputs. That allocation puts the following on the firm's side of the ledger.

  • Lawful basis. Contract with the client (Art. 6(1)(b)) covers processing needed to deliver the advice; legitimate interest (Art. 6(1)(f)) with the EDPB's three-step test covers the rest. Special-category data in employment, family, criminal and health-related files needs an Art. 9 ground, usually legal claims. "The client agreed" is rarely valid consent and never covers counterparties or witnesses.
  • DPIA. New technology plus sensitive data at scale meets Art. 35; agentic features raise the stakes, because ChatGPT Work takes an outcome and works "independently for hours" with access to 1,400+ plugins. The DPIA should also record sub-processors, which change: Microsoft's Legal Agent for Word (30 April 2026) requires Anthropic to be enabled as a sub-processor, a useful reminder that a vendor's AI stack is rarely one company.
  • Transfers. Residency settings answer where data sits, not who can compel it; Mistral's argument for EU domicile is that a "data-residency toggle changes where data sits — not which government can compel its production". Check the current processing locations in the contract and record the Chapter V mechanism.
  • Rights and records. A DSAR now covers prompts and outputs about the data subject; the Art. 30 record needs an entry; retention of AI logs should match the file retention schedule rather than the vendor's default.

The GDPR and AI tools page works through each of these with worked examples, and the vendor due-diligence questions cover the contract terms to insist on.

Professional secrecy is stricter than the GDPR

For lawyers in Germany and Austria the data-protection analysis is the second gate, not the first. The BRAK's Hinweise zum Einsatz von KI (December 2024) states that tools which store or forward client data for model training breach § 43a Abs. 2 BRAO, regardless of any GDPR paperwork. ÖRAK's four practice rules from its first KI-Tag on 19 March 2026 go further than the GDPR requires: client data stays on EU servers, AI use is disclosed to clients, the lawyer keeps final responsibility and no client data is used for training, with an information sheet referencing § 40 Abs 3 RL-BA 2015. The CCBE's guide of 2 October 2025 adds "inform clients and let them object". A consumer ChatGPT account fails all four ÖRAK rules; a Business workspace with training off and EU processing can pass them, which is why the Austrian and German guidance page treats configuration, not brand, as the question. UK lawyers face the same layering through the SRA's August 2026 warning notice and the Upper Tribunal's remark in UK v SSHD [2026] UKUT 81 (IAC) that pasting client letters into public AI tools puts them "in the public domain".

The seven-step compliance route

  1. Choose the tier. Business, Enterprise or API. Block consumer logins at policy and, where possible, network level: Thomson Reuters' 2026 survey found 34% of professionals using unsanctioned "shadow AI", and the confidentiality checklist explains why that is the real exposure.
  2. Sign the paper. Processor terms, sub-processor list, retention and training clauses, breach-notification timing. Read them against BRAK and ÖRAK if you practise in Germany or Austria.
  3. Configure. Confirm training is off, set workspace retention, leave GPT-6 Astra disabled until the DPIA covers it, allowlist plugins and connectors, decide who receives the Compliance API audit trail and for how long. The privacy-settings walkthrough shows each screen.
  4. Run the DPIA. Purposes, data categories, the EDPB upstream assessment, transfers, agentic features, residual risk, DPO sign-off. Re-run it for each new feature; the Astra for Law page shows how much changed in one announcement.
  5. Settle basis and notices. Update the privacy notice and the engagement letter; give clients the CCBE-style right to object; disclose AI use where ÖRAK or your own bar requires it.
  6. Prepare for rights and records. A DSAR procedure that reaches prompts and outputs, an Art. 30 entry, a retention rule aligned with file closure.
  7. Train the people. Article 4 of the AI Act applies to any organisation whose staff use tools like ChatGPT, per the Commission's FAQ updated 27 July 2026: no certificates or tests, but an internal training record. The Article 4 page has a syllabus.

Where this leaves you

A firm that stops at "we bought Enterprise" has bought a promise from OpenAI and documented nothing of its own. A firm that runs the seven steps once, files the DPIA and repeats step four whenever a feature changes can defend its ChatGPT use to a supervisory authority, a bar and a client in the same afternoon. Module 6 of the AI Lab for Lawyers course walks through the GDPR and AI Act workflow with a DPIA template, and the privacy-lawyer page covers advising clients on the same questions.

Frequently asked questions

Can I use ChatGPT Plus for client work under the GDPR?

No. As of September 2026 the Free, Plus and Pro plans train on conversations unless the user opts out, the terms are consumer terms rather than an Art. 28 processor contract, and the provider retains copies that courts have already ordered produced. Even with training switched off, you have no processor agreement, no sub-processor list and no retention commitment, so you cannot document the processing the GDPR requires.

Is ChatGPT Enterprise GDPR-compliant?

It can be, which is different from being compliant out of the box. Enterprise and Business tiers do not train on your data by default and support contractual zero data retention, but the firm still has to sign the processor terms, configure retention and plugins, run a DPIA, settle a lawful basis, update client notices and keep a training record. Skip those steps and the tier alone proves nothing.

Did the Italian court decide that ChatGPT complies with the GDPR?

No. The Court of Rome annulled the Garante's €15 million fine on 18 March 2026 because the Irish Data Protection Commission had become OpenAI's lead authority on 15 February 2024, so the Garante lacked competence under the one-stop-shop. The court did not examine lawful basis, transparency or age verification. The merits questions moved to Ireland and remain unanswered as of September 2026.

Do I need a DPIA before using ChatGPT in my law firm?

Almost always. Client files routinely contain special-category data, the technology is new, and agentic features act on documents at scale, which together meet the Art. 35 threshold. The DPIA should cover the tier and configuration, retention, sub-processors, transfers, the EDPB's expectation that deployers assess unlawful upstream training, and how data-subject rights will be honoured for inputs and outputs.

Does the EU AI Act add anything for a firm that just uses ChatGPT?

Yes, one duty. Article 4 requires organisations whose staff use tools like ChatGPT to take measures to support AI literacy, and the Commission's FAQ updated on 27 July 2026 confirms it applies to any such organisation, with no certificates or tests required but an internal training record expected. National market surveillance authorities have supervised it since 2 August 2026. Everything else in the Act targets providers or high-risk deployers.

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.

Reserve a seat · €200 Maven money-back guarantee · certificate of completion

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

Tools

ChatGPT for Lawyers: GPT-6 Astra, ChatGPT Work and What to Switch Off

ChatGPT for lawyers as of September 2026. Which plans train on your chats, the 20-million-log discovery order, GPT-6 Astra, ChatGPT Work and a firm setup.

8 min read

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

AI Rules for Austrian and German Lawyers: ÖRAK, BRAK and CCBE Guidance

ÖRAK's four AI rules and § 40 Abs 3 RL-BA bind Austrian lawyers; BRAK's § 43a BRAO line binds German ones; CCBE guides cover both. With a bilingual checklist.

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 the EU AI Act Apply to Law Firms? A Direct Answer

Yes. An EU law firm whose staff use ChatGPT, Claude or Copilot is a deployer under the EU AI Act. Article 4 literacy applies now, Article 50 since August 2026.

6 min read