AI Regulation Tracker / Courts and the profession

Colombia Constitutional Court Sets Binding Criteria for Judges Using AI

Decided August 2, 2024; still the leading standard. In judgment T-323/24, Colombia's Constitutional Court reviewed a case where a judge had used ChatGPT and laid down guiding criteria for how judges may, and may not, use artificial intelligence. The headline principle: generative AI can assist, but it cannot replace the judge's own reasoning.

By Anthony Guerriero · The Leveraged Years Regulatory Research · Published July 21, 2026 · Last updated July 21, 2026

The short version

In judgment T-323 of 2024, decided August 2, 2024, Colombia's Constitutional Court addressed a case in which a lower-court judge used ChatGPT while resolving a health-rights (tutela) matter. The Court did not find a due-process violation on those facts, because the judge had reached the decision independently and used the tool as a support after the fact rather than as the decision-maker. But it used the case to set out guiding criteria for judicial use of AI. The core rule is that generative AI may support administrative and preparatory tasks but cannot substitute the reasoning, discretion, and human judgment of the judge. The Court also stressed transparency about AI use, rigorous verification of AI output, protection of data, and continued human oversight, and it ordered the judicial authorities to develop guidance and training. This is a binding constitutional ruling, and it has become a reference point across Latin America for how courts should treat AI.

Regulatory Briefing

Instrument: Sentencia T-323 de 2024 (judgment T-323/24)
Authority: Corte Constitucional de Colombia (Constitutional Court)
Jurisdiction: Colombia; binding on the judiciary, with persuasive weight regionally
Bindingness: Binding. A ruling of the Constitutional Court reviewing a tutela action. It resolves the case and sets criteria the judicial branch must observe when using AI.
What Changed: Colombia now has an authoritative judicial standard: AI may assist but must not replace a judge's reasoning, and judges using AI must disclose it, verify its output, and keep human oversight.
Who This Affects: Judges and judicial staff in Colombia, and by extension litigators, court administrators, and anyone watching how courts globally set rules for AI in adjudication.
Current Status: Decided August 2, 2024; in force. The Court ordered the judicial authorities to issue implementing guidance and AI training.
Primary Source: Corte Constitucional de Colombia, Sentencia T-323 de 2024, corteconstitucional.gov.co
Last Verified: July 21, 2026

Colombia produced one of the first serious appellate answers to a question every court system is now facing: what happens when the judge uses ChatGPT. In judgment T-323 of 2024, decided on August 2, 2024, the Constitutional Court reviewed a case in which a circuit judge had used a generative AI tool while handling a tutela, Colombia's fast constitutional-rights action, in a matter involving a child with autism. Rather than treat the episode as a scandal or wave it through, the Court used it to write down how judges may work with AI. That is why the ruling has outlived its own facts and become a reference across the region.

On the specific facts, the Court did not find that the judge violated due process. The reason matters. The judge had already reached the decision through independent legal reasoning and used the AI tool as a complementary support, not as the source of the ruling. The Court drew a bright line between those two things. AI as a drafting or research aid behind a decision the judge actually made is one thing; AI as the entity making the decision is another and is not permitted.

What rules did the Court actually set?

The Court laid out a set of guiding criteria for judicial AI use, and the through-line is that automation cannot displace the human judge. It insisted on what it called the non-substitution of human rationality: the reasoning, discretion, and judgment of the judge cannot be handed to a model. Around that anchor, the Court set expectations that any professional will recognize as sound AI governance. Judges must be transparent about when and how they use AI. They must verify its output rigorously rather than trust it, precisely because these tools produce plausible text that is not always accurate. They must protect personal and sensitive data, guard against bias and error, and keep meaningful human oversight over anything the tool touches. The Court was blunt about the technology's limits, noting that with a language model "no siempre la siguiente palabra mas probable es la mas correcta o cierta," the next most probable word is not always the most correct or true.

Crucially, the Court did not stop at principles. It ordered the judicial authorities, including the body that administers the judiciary, to develop concrete guidance on AI use and to provide training and AI literacy for judges and staff. In other words, it treated this as a systemic governance problem to be solved with rules and education, not a one-off disciplinary matter.

Why does a Colombian ruling matter to US professionals?

It is not binding outside Colombia, and I want to be precise about that. A US judge or lawyer is governed by US rules, and the many US sanctions decisions over fabricated AI citations are the ones that bind here. But T-323/24 matters for two reasons. First, it is one of the clearest early statements of a principle that is now converging worldwide: AI can assist the administration of justice, but the accountable human decision-maker cannot be automated away. US courts, Brazilian judicial councils, and others are landing in the same place from different directions. Second, the criteria the Court articulated are a ready-made checklist for any professional building an AI policy in a high-stakes, high-accountability setting. Disclose use, verify output, protect data, keep a human in charge, and train your people. Swap "judge" for "attorney," "physician," or "executive" and the framework holds.

The practical read

For attorneys and anyone advising on AI governance, the lesson from Colombia is the same one the sanctions dockets keep teaching, just stated from the bench in a structured way. The tool is allowed. Unverified reliance on it is not. The person who signs the decision owns the decision. Build your AI use so that a human has genuinely reasoned through the output, can show they checked it, and can explain how the tool was used. Colombia put that expectation into a binding constitutional judgment nearly two years ago. The rest of the world has been catching up to it since.

Questions professionals are asking

Did Colombia's court ban judges from using AI?

No. Judgment T-323/24 allows judges to use AI as a support tool for administrative and preparatory tasks. What it prohibits is letting AI substitute the judge's own reasoning and judgment. On the facts before it, the Court found no due-process violation because the judge had decided independently and used the tool only as a complement.

What must a judge do when using AI under this ruling?

Be transparent about the AI use, verify the output rigorously rather than trust it, protect personal and sensitive data, guard against bias and error, and maintain meaningful human oversight. The Court framed non-substitution of human reasoning as the anchoring rule and ordered the judiciary to develop guidance and training.

Is the ruling binding, and where?

Yes, within Colombia. It is a Constitutional Court judgment reviewing a tutela action, so it resolves the case and sets criteria the Colombian judiciary must observe. Outside Colombia it is persuasive rather than binding, but it has become a widely cited reference on AI in adjudication.

Does this affect US lawyers or judges?

Not directly; US courts are governed by US rules, and it is US sanctions decisions on fabricated AI citations that bind here. But the Colombian criteria, disclose AI use, verify output, protect data, keep a human accountable, work as a practical template for AI policies in any high-accountability profession.

RELATED BRIEFINGS

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how any court ruling or professional rule applies to your situation with qualified counsel in the relevant jurisdiction.