AI Regulation Tracker / Courts and the judiciary
Colombia Binds Judges to Authorized AI Tools and Human Oversight
Adopted December 16, 2024 and in force. Colombia's Superior Judicial Council issued Agreement PCSJA24-12243, a binding set of rules for how magistrates, judges, and judicial staff may use generative AI. It requires human verification, bars automated evaluation of evidence, and demands disclosure whenever AI is used in judicial work.
Colombia has done what most court systems are still debating. On December 16, 2024, the Consejo Superior de la Judicatura, the body that administers the Colombian judiciary, adopted Agreement PCSJA24-12243, a binding set of rules for how everyone inside the Judicial Branch may use generative AI. It reaches magistrates, judges, judicial employees, and administrative staff across all jurisdictions and specialties. This is not a hortatory ethics memo. It is an administrative act, and inside the judiciary it functions as mandatory internal law.
The Agreement did not come out of nowhere. It responds directly to the Constitutional Court's ruling T-323 of 2024, the decision that pressed the judiciary to put responsible and safe AI use on a formal footing. So Colombia now has two things pointing the same direction at once: a constitutional court decision telling judges to be careful, and an operational rulebook telling them exactly how.
What does the Agreement actually require?
The core move is that AI cannot replace the judge. The Agreement prohibits using AI to automatically evaluate evidence and bars any use that would substitute the human judgment at the center of the judicial function. Around that hard line, it sets a tiered structure of permitted uses: routine administrative support, uses that require detailed human review, and ethically sensitive applications that demand the strictest compliance with transparency, accountability, and privacy duties.
Human oversight is not optional. The Agreement requires users to control and verify the reliability of any generative AI they use and to carry out a detailed verification of the products and results before relying on them. In plain terms, a judge or clerk who uses an AI tool owns the output completely and has to check it. The technology assists; it does not decide, and it does not get to be wrong on the record without someone accountable having signed off.
What has to be disclosed?
This is the part US practitioners should study most closely. The Agreement requires explicit disclosure of AI use, and the disclosure is specific. Users must identify the tool name, the model, the provider, the date of use, and the actual prompts employed. That is a far more granular disclosure standard than most US courts have adopted, where a general certification that AI was or was not used is still the norm. Colombia is asking for the receipts.
As the President of the Superior Judicial Council, Diana Alexandra Remolina Botia, framed it, the Agreement pairs "the first impulse to AI use with self-control, but also rules that must evolve as technology, normativity, and jurisprudence advance." That is an honest description of what this is: a first, binding framework that expects to be revised, not a finished code.
Why should a US attorney care about a Colombian judicial rule?
Two reasons. First, if you practice cross-border or have matters that touch Colombian courts, this is now the operating environment, and the disclosure and human-verification duties are real for the judges deciding your cases. Second, and more broadly, Colombia is a preview. The disclosure-plus-verification standard here, name the tool, name the model, verify the output, keep a human accountable, is the same shape US courts are moving toward through local standing orders and sanctions rulings, one bench at a time. Watching a national system codify it in one instrument tells you where the fragmented US approach is heading.
The practical read for US lawyers is not compliance today. It is calibration. If you want to know what a mature judicial AI-use rule looks like, this is one of the clearest in force. Build your own AI-use practices, disclosure habits, and verification workflows to a standard you would be comfortable defending under a rule this explicit, because some version of it is coming to a courtroom you appear in.
Questions professionals are asking
Is Agreement PCSJA24-12243 binding on Colombian judges?
Yes. It is an administrative act (acuerdo) of the Consejo Superior de la Judicatura and applies to magistrates, judges, judicial employees, and administrative staff across all jurisdictions. It is not a statute passed by Congress, but within the judiciary it is mandatory internal law adopted on December 16, 2024.
Can a Colombian judge use AI to weigh evidence?
No. The Agreement prohibits using AI to automatically evaluate evidence and bars uses that would substitute the judge's own decision. AI may support administrative and drafting work, but the human judgment at the center of the judicial function cannot be delegated to a model, and outputs must be verified by a person.
What must a judge disclose when AI is used?
The Agreement requires explicit disclosure that identifies the tool name, the model, the provider, the date of use, and the specific prompts used. That is a more granular standard than most US courts require, where a general AI-use certification is still typical.
Does this Colombian rule create any duty for US attorneys?
Not directly. It governs Colombia's judiciary. But it is instructive: it codifies a disclosure-plus-verification standard that US courts are approaching through local standing orders and sanctions rulings. US lawyers should treat it as a preview and calibrate their own AI-use and disclosure practices accordingly.
RELATED BRIEFINGS
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel in the relevant jurisdiction.