Les avocats peuvent-ils utiliser l'IA ? Un tracker pays par pays (2026)

The Short Answer

Can a lawyer use AI? In almost every major legal market, yes — provided they treat the output as a draft, not an oracle. The recurring theme across every regulator that has spoken is the same: AI does not dilute a lawyer's existing duties. You still owe the client competence, confidentiality, and a duty to verify. The tool changes; the responsibility does not.

Key facts

• Across 17 jurisdictions tracked: 8 permit lawyer AI use with formal guidance, 2 restrict or require disclosure (EU, Qatar), 7 have not addressed it.
• Qatar's QICDRC Practice Direction No. 1 of 2026 is MENA's first hard rule: AI-generated content must be flagged and verifiable on affidavit.

The American Bar Association set the template with Formal Opinion 512 in 2024: understand the tool's limits, protect client confidences, verify the output, and bill reasonably. Most other regulators that followed echo it. The interesting story is not the consensus — it is the map of where regulators have said nothing at all.

The Tracker: 17 Jurisdictions

We ran live searches across 17 jurisdictions and classified each as permitted with guidance, restricted or disclosure-required, or not yet formally addressed. Every classification traces to a regulator instrument or its documented absence.

The MENA Picture

Is it ethical for lawyers to use ChatGPT? The duty of competence

The question lawyers actually type into a search bar is narrower than 'can lawyers use AI.' It is 'is it legal for lawyers to use ChatGPT,' and the honest answer is: yes, but the bar moved the goalposts. In the United States, using a general-purpose chatbot is not prohibited. What is prohibited is using one badly. The American Bar Association did not write a new rule for AI; it pointed the existing Model Rules of Professional Conduct at it. ABA Formal Opinion 512, issued July 29, 2024, runs the analysis through Rule 1.1 (competence), Rule 1.6 (confidentiality), Rule 5.3 (supervision of non-lawyer assistance), Rule 3.3 (candor to the tribunal), and the reasonable-fee rules.

Competence is the one most lawyers underestimate. Opinion 512 says a lawyer must understand a tool's 'benefits and risks' before using it, and it adds a line most people skip: this is not a one-time box to tick. In the ABA's words, 'technological competence presupposes that lawyers remain vigilant about the tools' benefits and risks,' because the tools change underneath you. A lawyer who cannot explain, in plain terms, that a consumer chatbot generates plausible text rather than retrieving verified law is not yet competent to use it on a client matter. That is not our opinion. That is the framing the regulator chose.

By March 2026, more than 35 US state bar associations had issued their own AI guidance, and they echo the same spine: understand the tool, protect the client, verify the output, bill honestly. The pattern is identical to what every regulator in our tracker above arrived at. The tool is allowed. The duty is non-delegable. You can hand the typing to a model; you cannot hand it the responsibility.

The confidentiality risk: consumer ChatGPT vs a legal-grade tool

This is where 'can I use ChatGPT' quietly becomes 'should I.' The single hardest duty for a chatbot to satisfy is Rule 1.6 confidentiality. Opinion 512 is direct about it: because many self-learning AI tools are 'designed so that their output could lead directly or indirectly to the disclosure of information relating to the representation of a client,' a lawyer must get the client's informed consent before inputting that information into such a tool. Not a line buried in an engagement letter. The ABA specifically flags that boilerplate consent 'may not suffice.'

The reason is architectural, not legal. On consumer ChatGPT (Free and Plus), the default is that OpenAI may retain your prompts and use them to train its models unless you turn that off, and conversations are typically retained for a period regardless of the training toggle. The same is true of consumer-tier Claude and Gemini. Opting out of training does not erase the provider's right to retain data or disclose it in response to legal process. The moment client material lands on that infrastructure, you have arguably shared a privileged communication with a third party who owes your client nothing.

That risk is not theoretical. In a 2026 Southern District of New York ruling, a federal court held that documents a litigant generated with a consumer-tier AI tool and later shared with counsel were neither privileged nor protected as work product — an AI tool is not an attorney, holds no license, and owes no duty of confidence. We unpack that decision in our piece on why AI conversations are not privileged.

A legal-grade tool closes the gap by contract and by construction. Enterprise and commercial tiers (ChatGPT Enterprise, Claude's commercial plans) exclude your inputs from training by default and attach contractual confidentiality. Purpose-built legal platforms go further: zero-data-retention agreements with the model providers, deployment inside the privilege rather than outside it, and an audit trail of what the AI did and what a human approved. The distinction that matters is not 'AI vs no AI.' It is 'a tool you can give a confidentiality guarantee on vs a free consumer product where, when the tool is free, your client's data is the product.'

We tested what this looks like in practice: the same NDA review run on ChatGPT and a purpose-built legal AI, and where generic AI falls short of legal-specific tooling.

The Gulf, Levant, and North Africa skew heavily toward that silent group — with one sharp exception. Qatar issued the region's first hard rule: the QICDRC Practice Direction No. 1 of 2026 requires lawyers to flag AI-generated content and stand ready to verify it on affidavit. It emerged from an actual case, not a think-tank.

Elsewhere the groundwork is visible but unfinished. The UAE has a non-binding AI ethics charter; Saudi Arabia has SDAIA guidelines — neither aimed at practising lawyers. The Beirut Bar Association has stood up an AI committee and signed a 2026 government MoU. The Egyptian Bar ran a 'Generative AI for Lawyers' training event in late 2025. Real motion, no binding rule yet. For Morocco and Jordan we found awareness but nothing directed at lawyers.

For a region HAQQ is built for, that is the whole opportunity: in markets where the norm is not yet codified, the firms that adopt AI responsibly now will set the standard others get measured against.

What It Means

'Is it allowed?' is the wrong question to stop on. Every regulator that has spoken says the same thing in different words: you may use AI, and you remain fully responsible for the result. So the real question is whether your tools make that responsibility easy to discharge — citations you can check, confidentiality you can guarantee, a record of what the AI did and what a human approved.

That is the bet we made with HAQQ: build for the duty, not around it. When the Gulf regulators do write their rules — and Qatar shows they will — tools designed for verification and disclosure will already be compliant. The rest will be scrambling.

Key Takeaways

• Across 17 jurisdictions: 8 permit with guidance, 2 restrict/disclose, 7 unaddressed.
• Every regulator agrees AI does not reduce a lawyer's duties.
• Qatar is MENA's first hard rule; the rest of the Gulf is still silent.
• Early responsible adoption sets the regional standard.

Sources & Further Reading

• 1,458 court cases with AI-fabricated citations
• the sanctions record: 496 attorneys, $55K in fines
• our jargon-free LLM guide for lawyers
• ABA Formal Opinion 512 — generative AI and ethics rules
• EU AI Act — regulatory framework
• Qatar QICDRC Practice Direction on AI in proceedings
• BRAK (Germany) — AI guidance for law firms