Attorney-Client Privilege and AI Tools: What Every Legal Professional Should Know

The question that gets asked is “does using AI waive privilege?” It is the wrong question in most of this region, and it is the wrong shape of question even where privilege applies.

Privilege is not the same thing as professional secrecy

Legal professional privilege, in common-law systems, is a right belonging to the client. It allows the client to resist compelled production of communications made for the purpose of obtaining legal advice. It is an evidentiary shield, raised against a court or tribunal, and it can be lost — waived — by the client's own conduct.

Professional secrecy, the civil-law mechanism that most GCC jurisdictions use, is a duty imposed on the lawyer. It is enforced by the bar or the ministry rather than raised as an objection to disclosure, it can carry disciplinary and in some states criminal consequences, and the client generally cannot simply hand it back.

The practical difference matters. Under a privilege regime, the question after an unwise disclosure is “has the shield been lost?” Under a secrecy regime, the question is “has the lawyer breached a duty?” — and the answer to that does not depend on whether anyone ever tried to compel production.

In arbitration, the tribunal decides

International arbitration does not import one national law of privilege. Where the parties have adopted the IBA Rules on the Taking of Evidence in International Arbitration (2020), Article 9.2(b) lets the tribunal exclude material on grounds of legal impediment or privilege under the rules it determines to be applicable.

Article 9.3 then lists what the tribunal may weigh. Two entries on that list should hold a legal team's attention:

• The expectations of the parties and their advisors at the time the privilege is said to have arisen. A document handled as though it were confidential supports the expectation; one pasted into a consumer chat product does not.
• Any possible waiver by virtue of consent, earlier disclosure, or affirmation. “Earlier disclosure” is the phrase that does the work here. Sending material to an outside service is a disclosure, and if it becomes known, it is a fact the tribunal is entitled to weigh.

None of this means a tribunal will treat use of a contracted, confidential processing tool as a waiver. It means the tribunal will look at how the material was actually handled — and that “how” needs to be describable.

The disclosure that is hard to explain

Set two scenarios side by side. In the first, a firm engages a vendor under a written processing agreement with defined retention, no training on customer content, and a stated hosting region. In the second, an associate pastes a submission into a personal account on a consumer product whose terms they have not read.

Both are disclosures to a third party. Only one can be described to a tribunal, a regulator, or a client without the description itself being the problem. That is the whole distinction, and it is a procurement decision rather than a technology one.

What to actually do

1. 01Work out, per matter, which forum's rules govern — onshore court, DIFC or ADGM, or arbitral tribunal. The answer changes what you are protecting and against whom.
2. 02Treat every AI tool as a disclosure recipient and paper it accordingly, before anything confidential is processed rather than after.
3. 03Record what was processed and under which agreement, at the matter level. You are unlikely to be asked; if you are asked, nothing else will do.
4. 04Check outside-counsel guidelines and engagement letters. Some already require notice or consent before AI processing, which converts an ethics question into a contract question.
5. 05Keep consumer accounts out of it entirely. That single rule removes most of the exposure discussed here.

The confidentiality and data protection side of this — professional secrecy, the GCC data protection statutes, and the GDPR — is covered in why pasting case files into ChatGPT could violate client confidentiality.