Jamaica has had a binding generative AI practice direction in force since September 2025, and it covers confidentiality as well as verification

Why we are covering an eleven-month-old rule

Jamaica issued this in September 2025. It has been in force ever since. We are running it now because the US legal press has covered the sanctions cases exhaustively and the standing rules of foreign courts almost not at all, and because a rule that already binds is a more useful thing to know about than a draft that does not.

So treat what follows as a description of current law in Jamaica, not as a report of a new development. If your firm has a matter in the Jamaican Supreme Court, the Revenue Court, or either division of the Gun Court covered by paragraph 1.1, the obligations below applied to your last filing as much as to your next one.

The instrument is short. Six pages, nine parts, and an Appendix containing a declaration form. It is worth reading in full rather than through anyone's summary, including ours.

Who it binds, and who it does not

The opening recital states that the direction sets out guidance on the appropriate use of Generative Artificial Intelligence tools by attorneys-at-law, parties, self-represented litigants, and other persons involved in proceedings before the Court. Paragraph 5 places the due diligence obligations specifically on attorneys-at-law and self-represented litigants.

One point of precision, because the question comes up. We did not find a provision directed at court staff or registry personnel. Judicial officers appear once, at paragraph 8.1, and there the direction addresses what they may consider rather than what they may use: judicial officers may consider AI-assisted materials submitted by parties only to the extent that such materials meet the evidentiary and procedural standards of the Court. Anyone briefing a client that this rule governs Jamaican judges' own use of AI is reading in something the text does not say.

Paragraph 2.5 defines who counts as a party, by reference to whether their presence is necessary for complete adjudication and whether they hold a direct legal interest in the subject matter. That matters for instructing firms: the obligation attaches to the party as well as to counsel.

The verification duty and the declaration

Paragraph 3.1 is the sentence everyone quotes, and it says that court documents generated in whole or in part with the assistance of GenAI "must be thoroughly reviewed, verified, and approved by the party or attorney submitting them." Paragraph 3.2 then breaks that down: the document is to be fact-checked and legally accurate, and is not to contain fabricated case law, misapplied authorities, or hallucinations. Paragraph 3.3 adds that attorneys and parties are not to rely solely on GenAI for legal analysis or citation without confirming the validity of the authorities relied upon.

Disclosure is required, and it is required in a specific form. Under paragraph 6.1, where any part of a submitted document was prepared using GenAI, the party or attorney is to include a declaration in the form set out in the Appendix. The Appendix form names the tool and its version, and carries a confirmation that all information, citations, and representations in the identified portions have been independently verified by the person signing.

Read those two together and the practical shape is clear enough. The declaration is not a checkbox. It identifies which portions of the document were machine-assisted, and it is signed. Paragraph 6.2 lets the Court direct a party to provide further particulars of their GenAI use in any case, which means the declaration is a starting point for inquiry rather than the end of one.

Paragraph 5.2 forecloses the obvious argument in advance: use of GenAI tools does not relieve the party or attorney of their duty of candour and diligence to the Court.

The confidentiality limb, which is the underrated part

Paragraph 3.2(iii) provides that a GenAI-assisted court document is not to include confidential or privileged material submitted to unsecured platforms. Paragraph 5.1(ii) supports it from the other direction, requiring attorneys and self-represented litigants to exercise due diligence in ensuring the confidentiality of all data processed using GenAI tools. Paragraph 1.3 states protection of confidential and privileged information as one of the three purposes of the whole instrument.

That framing is worth sitting with. A verification rule polices your output. A confidentiality rule polices your input. An attorney who pastes a privileged memorandum into a consumer chatbot to get a summary has arguably engaged the second even if the resulting filing is flawless, every citation checks out, and no hallucination ever reaches the Court.

Our reading, offered as inference rather than as the Court's position: the direction does not define "unsecured platform", and it does not set out a technical standard for what makes a platform secure. Firms will have to make that judgement themselves, and document how they made it. The absence of a definition is the thing to plan around.

For US readers the analogy is not the hallucinated-citation sanctions docket. It is the confidentiality opinions the state bars have been issuing about generative AI tools. Jamaica has put both concerns inside one binding court rule, which is less common than it sounds.

Prohibited uses and stated consequences

Part 4 lists seven prohibitions. Among them: drafting or altering affidavits or witness statements intended to reflect a person's direct knowledge, belief, or opinion; generating expert evidence or any report purporting to be from an independent person; creating or modifying material intended to be tendered into evidence unless authorized by the Court; and generating documents purporting to be original documents.

Two of the seven concern images. Paragraph 4.1(v) restricts generating images unless expressly requested, explained and approved by the Court. Paragraph 4.1(vii) is narrower and more specific: it addresses generated images or avatars that falsely present or imply an accurate likeness of the user or any other person for the purpose of joining virtual conferences, meetings or hearings before Judges, Masters, Registrars, and Deputy Registrars. That provision does not appear in most of the equivalent rules we have read.

On consequences, the text is explicit and we will not stretch it. Paragraph 7.1 gives the Court three responses where a party or attorney submits inaccurate, fictitious, or misleading GenAI-generated information: refuse to admit the document into evidence, disregard the submission in part or in whole, or make any order as to costs or other sanction as may be appropriate. Paragraph 7.2 goes further, providing that non-compliance may constitute contempt of court, result in referral of attorneys-at-law to the Disciplinary Committee of the General Legal Council, or lead to any other sanction the Court considers just.

Note the modal. Paragraph 7.1 says the Court may. These are available responses, not automatic ones, and the direction fixes no tariff, no presumptive penalty, and no minimum sanction. We have seen no reported Jamaican decision applying the direction and we do not assert that one exists.

What we did not verify

What we opened: the full six-page PDF of Practice Direction No. 1 of 2025 on the Supreme Court of Judicature of Jamaica website, at the URL in the primary links above. Every paragraph number, every quotation, the 17 September 2025 date, the signature block naming Chief Justice Bryan Sykes, and the Appendix declaration form were read in that document. The quoted sentence at paragraph 3.1 was character-matched against the extracted text.

What we did not open: the practice directions of the other Caribbean courts. Secondary commentary asserts that six Caribbean courts now have generative AI rules, naming the Caribbean Court of Justice, Jamaica, Turks and Caicos, The Bahamas, Belize and Trinidad and Tobago. We opened only Jamaica's. We did not read the CCJ, Bahamian, Belizean, Turks and Caicos or Trinidadian instruments for this piece, so we make no claim about what any of them says, whether they are drafted alike, or whether the count of six is right. We also did not open any Jamaican court judgment applying this direction, and did not search the Jamaican cause list.

What we refuse to claim: that any specific sanction follows from a breach. Part 7 sets out what the Court may do, and we will not convert that into a prediction. We do not claim the direction applies to the Court of Appeal of Jamaica, to the Parish Courts, or to any court outside the four named in paragraph 1.1, because the text names those four and we did not verify the position elsewhere. We do not claim it governs judicial officers' own use of AI tools. And we do not offer a view on what counts as an unsecured platform, because the instrument does not define it and we are not going to invent a standard the Chief Justice did not write.

Related briefings

• Trinidad and Tobago judiciary AI practice direction
• Qatar QICDRC AI court practice direction
• Ireland High Court HC142 on generative AI in court documents
• AI and privilege: the Heppner ruling for attorneys
• AI Regulation News hub