A lawyer in Lebanon asked us this before we'd even finished our demo: "Who is responsible for the case law the AI gives? We heard about a lawyer who filed ChatGPT citations that turned out to be fake."
It's the right question to ask before choosing any tool. The answer, everywhere a court or regulator has spoken, is the same.
The short answer
- The lawyer who signs is responsible. Not the tool, not the vendor, not the junior who ran the prompt.
- Courts are already sanctioning it. New York in 2023, Abu Dhabi in 2025, and more than 2,200 decisions worldwide by October 2026.
- Gulf regulators have put it in writing. The UAE Ministry of Justice's guiding principles for law firms (September 2026) say professional responsibility remains when technology is used. Qatar and DIFC courts require verification against sources.
- Where there's no AI-specific rule, the old duties still apply. Duty of competence, duty of candour to the court, duty of care to the client.
What the courts have done
| Court | Case | What happened |
|---|---|---|
| US District Court, S.D.N.Y. | *Mata v. Avianca*, 22 June 2023 | Lawyers filed a brief citing cases generated by ChatGPT that didn't exist, then defended them. $5,000 sanction, plus letters to the judges falsely named as authors. |
| ADGM Courts (Abu Dhabi) | *Arabyads v Alam*, December 2025 | Defence relied on non-existent or misapplied authorities. AED 282,508 in wasted costs ordered against the former legal representatives. |
| Worldwide | AI Hallucination Cases database | 2,202 decisions involving AI-fabricated citations or content, as of 19 October 2026. |
The pattern in these cases is telling. The sanction is rarely for using AI. It's for not checking, and, in Mata, for doubling down when challenged.
What the rules say in the region
| Body | What it says | Source |
|---|---|---|
| UAE Ministry of Justice, 29 Sep 2026 | Guiding principles for AI in law firms. Professional responsibility remains when using technology; AI outputs need careful human review and verification. | |
| QICDRC (Qatar), PD 1/2026 | All citations must be checked against primary sources or recognised databases. Sanctions include referral to regulators. | |
| DIFC Courts, PGN 2/2023 | Don't rely on AI content without first verifying it; check the sources the tool relied on. | |
| SRA (England and Wales), 17 Aug 2026 | AI does not diminish or transfer your professional responsibilities. |
In Lebanon, Egypt and Kuwait we did not find a published AI-specific rule for lawyers as of October 2026. That doesn't create a gap. A lawyer who files a fake authority there still faces the general rules on professional conduct and duties to the court.
Can you pass the risk to the vendor?
Mostly no. Your engagement letter is with your client, and your duty to the court is personal. A vendor contract may give you a claim against the vendor in some cases, but it won't undo a costs order or a disciplinary file. The practical protection is a verification step, not a contract clause.
What to put in place
- A verification rule. Every authority checked at the source before filing. See How to Check an AI Legal Citation Before You File It.
- Separate maker and checker. The person who prompted the tool shouldn't sign off alone. See The Maker Cannot Be the Checker.
- A note in the file. Which tool, which task, which sources were opened, who checked.
- Client consent for confidential inputs, where your rules or your client's instructions require it.
- A rule for when you're challenged. If an authority is questioned, check it immediately and correct the record. Mata became a sanctions case because the lawyers didn't.
Where HAQQ Legal AI fits
HAQQ Legal AI shows a reference beside each paragraph so verification takes a click instead of a search. That makes the checking step realistic on a busy day. It doesn't move the responsibility, and we'd be wary of any vendor that says it does.
Încercați HAQQ AI gratuit
Experimentați redactarea juridică și cercetarea asistate de AI
Frequently asked questions
If a junior uses AI and makes the mistake, who's responsible?
The supervising lawyer who signs, and the firm. Supervision duties apply to AI-assisted work the same way they apply to any delegated work.
Is using AI itself a breach of duty?
No regulator we've cited bans it. The duties are to use it competently, protect confidentiality, and verify.
Does disclosure to the court protect me?
It may be required, depending on the court. It doesn't excuse a fake citation.
Sources and further reading
- HAQQ Legal AI, "AI Hallucination Cases: The 2,202-Case Sanctions Tracker", based on Damien Charlotin's database (last updated 19 October 2026)
- ABA Journal, "Lawyers who 'doubled down' and defended ChatGPT's fake cases must pay $5K, judge says", June 2023
- Bloomberg Law, 22 June 2023
- Oxford Institute of Technology and Justice, UAE AI Justice Atlas, March 2026
- Matouk Bassiouny & Hennawy via Legal 500
- Emarat Al Youm, "Minister of Justice launches guiding principles for the use of AI systems in law firms", 29 September 2026
- QICDRC Practice Direction No. 1 of 2026, as analysed by HAQQ Legal AI
- DIFC Courts, Practical Guidance Note No. 2 of 2023
- SRA, "Warning notice: misuse of AI", 17 August 2026
