Colombia | increasing AI use in law enforcement, prosecution, and courts | Oxford Institute of Technology and Justice

Tools

Tasks

Case management | Charging support | Data review and analysis | Legal research, analysis and drafting support | Predictive analytics

Users

Law enforcement | Prosecutors | Courts

Scope

Nationwide

Training

Not mandatory or systematic

Regulation

No dedicated legislation targeting the use of AI in criminal proceedings, but the Superior Council of the Judiciary issued guidelines on the use of AI. Existing laws regulating criminal procedure and data protection also apply

Cases

The Constitutional Court ruled that the use of ChatGPT in drafting a judicial decision is permissible so long as it does not replace human reasoning. In another ruling, the Constitutional Court held that by refusing to disclose the source code of a public health app, the authorities violated the claimant’s fundamental right of access to public information, emphasising the importance of algorithmic transparency

Insights

PretorIA, implemented by the Colombian Constitutional Court to assist in docket selection for tutela (constitutional protection action) cases, processes approximately 4,500 cases daily

AT A GLANCE

In Colombia, AI is used across law enforcement and the judiciary to improve efficiency and decision-making. The National Police use predictive analytics with Amazon Web Services and Nuvu to forecast crime and improve emergency response, while facial recognition systems in Bogotá and Medellín aid in identifying fugitives and Clearview AI has been trialed for child-exploitation cases. For prosecutors, tools like Fiscal Watson analyse millions of complaints to detect patterns, while PRiSMA predicts recidivism risk to guide pretrial detention decisions, and PROMETEA (originally from Argentina) accelerates case processing with high accuracy. In the courts, the PretorIA system helps the Constitutional Court manage the huge volume of tutela cases, acting as a search engine to highlight important filings, while judges also experiment with generative AI for drafting and summarisation. Training remains uneven, with efforts underway but no fully structured national programme yet.

Despite this scope of use, as at August 2025, there is no national legislation that expressly regulates the use of AI in criminal proceedings or court proceedings more generally. However, in December 2024, the Superior Council of the Judiciary (Consejo Superior de la Judicatura) adopted Guidelines for the Responsible and Safe Use of Generative Al in the Judicial Branch. This made Colombia the first country to implement the UNESCO Draft Guidelines for the Use of AI Systems in Courts and Tribunals. The Guidelines establish a general framework for generative AI use within the judiciary, emphasising that AI must not replace human reasoning and that explainability must be ensured. AI-generated outcomes must be verified and their use must be disclosed and documented with a detailed record. The Guidelines followed the August 2024 ruling of Colombia’s Constitutional Court, which approved the use of ChatGPT by a judge on the basis that it did not replace judicial decision-making.

USE

Law enforcement

Predictive analytics

The National Police of Colombia partnered with local innovation firm Nuvu and Amazon Web Services to develop an AI-powered incident prediction and response system. The platform ingests and processes large amounts of historical and real-time crime data, including 911 emergency calls, geospatial information, and patterns of criminal activity across cities. Using machine learning algorithms, the system generates predictive models that help anticipate where and when incidents are most likely to occur.

Data review and analysis

Facial recognition tools are used in urban areas in Colombia:

In Bogotá, as part of a 30-day proof of concept at the TransMilenio bus system, Corsight AI’s facial recognition was deployed across 20 cameras. In just two weeks, authorities identified and apprehended six individuals (one for homicide, five for theft), matching images against a database of over 5,000 with active court orders.
In Medellín, the Colombia National Police selected Herta’s facial recognition software as part of the efforts to strengthen public security. Herta deployed 80 facial recognition cameras in public transport and key locations using Herta’s technology. These systems monitor roughly 19,000 persons with current arrest warrants.

In March 2024, law enforcement units in Colombia took part in a five-day trial of a facial-recognition tool for online child-exploitation cases, developed by American company Clearview AI. The tool allows law enforcement units to upload images and run them through a database of billions of public photos from the Internet. In combination with participants from nine other countries, the tool was used on a total of 2,198 images and 995 videos, hundreds of them from cold cases. In just three days, they identified 29 offenders and 110 victims. By June 2024, at least 51 victims had been rescued as a result of the effort.

Prosecutors

Case management

‘Fiscal Watson’, developed by the Prosecutor General’s Office and announced in 2018, is an AI system that analyses Colombia’s extensive criminal database, encompassing over 13 million complaints from citizens or public officials since 2005. The tool assists prosecutors and investigative staff by linking vast sets of data and identifying patterns such as modus operandi, physical characteristics, and recurring use of weapons or vehicles used by suspects. Fiscal Watson functions primarily as a search platform that makes it possible to search for complaints with similar characteristics using key words in the statement of facts for crime reports logged in the ‘Oral Accusatory Criminal System’ (SPOA). The tool can also match cases based on quantitative, qualitative, and geographical components, using data relating to the district attorney’s office, the date of the incident, and its location. Fiscal Watson is used, in theory, at the investigation stage to search for and correlate investigations that meet established criteria, which may be geographical or qualitative variables that are found in the statement of facts logged in the SPOA. The tool indexes large sets of structured, unstructured and semi-structured data from disparate data sources, and applies text analysis mechanisms to find relationships, generate warnings, and automatically extract indices and general contextual information.

Charging support

'PROMETEA', launched by the Buenos Aires’ prosecutor’s office, has, since 2019, seen increasing use in Colombia to predict judicial rulings. PROMETEA was adapted to Colombia’s legal and linguistic requirements by the University of Buenos Aires Innovation Laboratory with the Colombian judiciary. It is linked to the Colombian courts’ case-management and document-filing systems, enabling secure data exchange. PROMETEA uses natural-language processing to scan tutela petitions and process court rulings in less than 20 seconds with an accuracy rate of 96% in identifying relevant cases. In the Colombian Constitutional Court, itreduces the time allocated to the selection of urgent cases from 96 days to two minutes. While PROMETEA is a major efficiency tool in case triage, final decisions on case selection remain with human judges.

In 2016, the Colombian Attorney General’s Office initiated ‘PRiSMA’ (Perfil de Riesgo de Reincidencia para Solicitud de Medidas de Aseguramiento), a program operated through AI which aims to determine the risk of recidivism of persons accused of crimes to support the decision of the Prosecutor General’s Office on whether or not to request preventative detention for someone under investigation by Colombian authorities. PRiSMA is based on an algorithm that uses machine learning to process information from a database provided with information by the National Police and Attorney General’s Office, which has information on six million individuals with criminal records. From this, PRiSMA predicts the probability that a person would reoffend. The tool and algorithms are implemented through a PDF document that prosecutors can download from the SPOA when requesting a restraining order. This document contains all the information available on the subject: the number of previous police arrests, SPOA and court proceedings, and all information on prior incarceration events. If, using this data, the system predicts that the person represents a high risk of re-offending, the Prosecutor General's Office can request a more restrictive measure. If the system predicts a low risk, non-restrictive pre-trial preventative measures are requested. The final legal decision remains with human actors, and the system is intended to serve as an auxiliary tool only. PRiSMA was originally tested by a smaller group of ten prosecutors, and gradually expanded.

Courts

Case management

‘PretorIA’, implemented by the Colombian Constitutional Court in 2020 and inspired by PROMETEA (in the Buenos Aires Attorney General’s Office), is an AI mechanism designed to assist in docket selection for tutela cases, which are constitutional protection actions and also enables the preparation of statistics and the identification of recurring issues. In Colombia, the Constitutional Court receives more than 600,000 tutelas per year.

PretorIA aids this procedure by processing approximately 4,500 cases daily, identifying patterns, generating statistics and organising workload, but each case is still reviewed individually, and PrestorIA functions as a trained search engine rather than a decision-maker.

Though the mechanism was originally intended to act as a decision-making tool, it now acts as a trained search engine to aid judges and court clerks in identifying important cases from the large volume of tutela filings, but retaining the role of the human judge as the sole decision-maker. There is, however, potential for the expansion of its capabilities. The implementation of PretorIA faced challenges such as digitising paper-based files and refining the algorithm, but is now used in tandem with the traditional case-by-case review process.

PROMETEA, discussed above, is also used in Colombia to assist in the selection of tutela petitions by automatically reading large volumes of filings and flagging urgent cases for priority review, but the Constitutional Court retains discretion to confirm, modify, or reject the system’s suggestions.

Legal research, analysis and drafting support

Judges report unofficial use of commercial AI tools such as ChatGPT and CoPilot for drafting and summarisation tasks.

Colombian courts are exploring additional applications of AI, including the use of generative AI to summarise public interventions in constitutional cases, and the introduction of a ‘cognitive file’ system that would allow judges to query case information in natural language. There is also an initiative to improve jurisprudence search capabilities using AI-assisted mapping of related cases.

Colombian Judge, July 2025

Defence

There are no reported cases of defence counsel in Colombia making use of AI.

TRAINING

Colombia’s judiciary has introduced training initiatives such as a ‘competencies portal’ as part of its AI strategy, aimed at training judges at different levels, from basic digital awareness to advanced technological development. Innovation contests and stakeholder dialogues have also been organized to encourage bottom-up innovation proposals for applying technology in judicial work. However, there remains a lack of structured training for judges and court staff on how to use AI tools appropriately, including on crafting effective AI prompts and ethical limitations on the use of AI.

REGULATION

As at August 2025, there are no national laws or statutory rules that expressly regulate the use of AI in criminal proceedings or court proceedings more generally. However, binding judicial guidelines exist (Acuerdo No. PCSJA24-12243). Colombia has also adopted a national AI policy and the government in 2025 has introduced Bill No. 422 proposing a comprehensive AI regulation which is under consideration before Congress.

Guidelines for practitioners

Judicial guidelines

The Superior Council of the Judiciary (Consejo Superior de la Judicatura) – the body responsible for the administration and oversight of Colombia’s judicial system – has been promoting the use of AI ‘within the framework of the digital transformation in the Judicial Branch, with the aim of improving attributes of the administration of justice’. In December 2024, following a Colombian Constitutional Court ruling and a survey on the use of AI in the judiciary, the Superior Council of the Judiciary adopted Guidelines for the Responsible and Safe Use of Generative AI in the Judicial Branch (‘Guidelines’) through Acuerdo No. PCSJA24-12243, ‘in order to maximise the benefits and potential of [AI tools], while mitigating and managing their potential risks’. This made Colombia the first country to implement the UNESCO Draft Guidelines for the Use of AI Systems in Courts and Tribunals. An unofficial English translation of the Guidelines is available here.

The Guidelines apply across the entire judicial branch, including national and local courts, judges, clerks, and administrative staff. They include categories of permitted use with exhaustive lists of permissible tasks. Using AI to perform administrative functions– such as ‘scheduling of activities or errands’, drafting minutes or administrative texts and translating documents – is always permitted. When AI is used for more substantive tasks – such as transcribing hearings, summarising the facts of a case and drafting procedural orders – it is subject to additional safeguards and requirements, including – depending on the task – ‘detailed review’ of the output and ‘special observance of transparency and accountability’. Generative AI may not be used for any task outside those listed in the Guidelines, unless expressly authorised by the Superior Council of the Judiciary.

The Guidelines impose restrictions on both the use of certain AI tools and the type of tasks they may be used to perform:

Training Data Provenance: AI tools whose training data provenance cannot be verified may only be used to perform administrative support functions.
Evidence and legal reasoning: AI may not be used for evidence assessment, scrutiny of facts, making ‘value judgements’, and resolving ‘legal problems’.
Automated decision-making: AI tools that apply legal rules or adopt decisions based exclusively on the tool’s responses are prohibited.
Human rights risk: AI tools whose use could infringe fundamental rights are restricted.
Commercial free chatbots: General-purpose or commercial chatbots in their free version may not be used for judicial purposes.

The Guidelines set out a catalogue of 15 principles and guarantees governing the use of AI in the judiciary, including:

Primacy of fundamental rights	The judiciary must ‘ensure the respect, protection and promotion of fundamental rights’ when deploying and using AI and avoid ‘any form of discrimination’ and ‘any negative impact on human rights’
Human oversight, risk mitigation and prevention	Judicial officials and employees are responsible for exercising ‘strict scrutiny’ over any actions and decisions involving AI, including assessing ‘the sources, scope, restrictions, possibilities, shortcomings, and risks presented by the tool’ and checking for hallucinations, inaccuracies, and biases
Non-substitution of human rationality	AI tools may assist but must not replace judicial reasoning, the assessment of fact, the analysis of evidence, the application or interpretation of law, or decision-making
Transparency	Judicial officials and employees must disclose and document the use of AI, including details such as the tool name, provider, date of use, prompts, and outputs. AI-generated text must be clearly labelled in case files and judicial decisions
Explainability	The judiciary may only use AI tools that ‘provide evidence, reasons or justifications for the processes they perform and the results they generate, which must also be understandable to internal and external users’
Data protection, privacy and security	No sensitive or confidential data may be entered into AI tools that retain, share or use it to train their models. The judicial branch must ensure the integrity, confidentiality, and availability of the information managed through the AI system
Continuous monitoring	The Digital Transformation and IT Unit of the Superior Council of the Judiciary is responsible for overseeing the institutional deployment of generative AI systems

Criminal procedure rules

The Criminal Procedure Code (Law 906 of 2004) and, in transitional cases, the older inquisitorial procedure under Law 600 of 2000, establish admissibility, chain of custody, and authenticity standards applicable to AI-generated or analysed evidence. While these codes predate artificial intelligence and do not refer to it expressly, their provisions on digital and expert evidence apply to outputs generated or processed by AI systems (such as facial recognition matches, predictive analytics, or PRiSMA risk assessments) when introduced in criminal proceedings. In practice, such evidence would need to comply with general standards of authenticity and reliability and may require expert testimony to support its probative value.

Additionally, provisions of the Cybercrime and Technology Misuse Laws (Law 1273 of 2009 which amends the Penal Code to criminalise unauthorised access, data interception, and damage to information systems) apply to malicious uses of AI that compromise evidence or judicial data in criminal cases.

Data protection legislation

Law 1581 of 2012, also known as General Data Protection Law, and its regulatory decrees (Decree 1377 of 2013 and Decree 1081 of 2015), governs the protection of personal data in both the public and private sectors. Although the statute does not expressly mention AI, its provisions apply broadly to any ‘system’ that processes personal data. By extension, the requirements of Law 1581, including lawfulness, purpose limitation, security, and transparency, also govern the AI-assisted processing of personal data in judicial proceedings and criminal investigations conducted by courts and law enforcement.

The law is enforced by the Superintendence of Industry and Commerce (SIC), which can investigate and sanction improper AI-related data processing conducted by both public and private entities. The SIC issued Circular Externa 002 of 2024, clarifying privacy obligations for AI systems processing personal data, requiring privacy and risk-management safeguards for such systems, including those used in law enforcement and judicial contexts.

Human rights

The use of AI in criminal proceedings must be consistent with procedural guarantees and fundamental rights in the Colombian Constitution, including the right to due process and equality before the law. The Constitutional Court reaffirmed in its August 2024 ruling (T-323/2024), that AI tools cannot substitute judicial reasoning and must not undermine these protections. The Court required that whenever AI is employed in adjudication, its use requires disclosure, transparency, and human accountability.

Relevant guarantees, including the right to a fair trial and privacy rights, under other international human rights treaties to which Colombia is a party, such as articles 8 and 11 of the American Human Rights Convention, articles 14 and 17 of the International Covenant on Civil and Political Rights or articles 16 and 40 of the Convention on the Rights of the Child, may also provide additional guidance.

Outlook

In February 2025, the national government approved its first comprehensive National Artificial Intelligence Policy (CONPES 4144). The policy is designed to promote the development, adoption, and ethical and sustainable use of AI across both the public and private sectors, drawing on international guidelines such as the OECD AI Principles and the UNESCO Recommendations on the Ethics of Artificial Intelligence.

In June 2025, Colombia advanced its AI regulatory agenda by introducing Bill No. 422 inspired by the EU AI Act. The draft bill proposes a risk-based classification of AI systems, creation of a national oversight authority (under the Ministry of Science, Technology and Innovation), regulatory sandboxes, and binding principles such as human oversight, transparency, and privacy protection, along with sanctions for non-compliance. AI uses in law enforcement and the administration of justice are classified as ‘high-risk’. The draft also contemplates civil and criminal liability for harm from AI misuse. The Ministry of Justice and the Ministry of Science, Technology and Innovation have signalled that the justice sector will be prioritised for pilot projects under the forthcoming national AI bill.

Moreover, the Superior Council of the Judiciary is developing secondary regulations to operationalise the Guidelines for the Responsible and Safe Use of Generative AI in the Judicial Branch (Agreement PCSJA24-12243), including specific technical standards for approved AI tools.

CASES

The Colombian Constitutional Court assessed the legality of AI-assisted drafting of judicial decisions and ruled on whether citizens have the right to access the algorithms used in public administration.

In its August 2024 ruling (T-323/2024), the Constitutional Court addressed for the first time the use of AI by a judge. It found that the use of AI to draft a judicial decision did not violate due process rights, as the tool did not replace judicial decision-making. In the case at issue, AI merely supported the reasoning after the judge had already reached a decision. The Constitutional Court called on the Superior Council of the Judiciary to establish guidelines and, in the same ruling, set out guiding principles and criteria for the use of AI in the judiciary, including: (i) transparency; (ii) accountability; (iii) privacy; (iv) non-substitution of human rationality; (v) seriousness and verification; (vi) risk prevention; (vii) equality and equity; (viii) human control; (ix) ethical regulation; (x) compliance with good practices and collective standards; (xi) continuous monitoring and adaptation; and (xii) suitability.

In another ruling (T-067/2025), the Constitutional Court held in February 2025 that by refusing to disclose the source code of a public health application (‘CoronApp’), the authorities violated the claimant’s fundamental right of access to public information as protected by the Colombian Constitution and Ley No. 1712 of 2014 (Law of Transparency and Access to Public Information). The Court ordered the Ministry of Health to provide the source code and its version history, after taking necessary technical measures to protect personal data. It further instructed the government to develop guidelines for algorithmic transparency in public-sector digital tools.

The Court recognised that the source code of a public application such as CoronApp constitutes public information, particularly when the application is used to implement public policy and affects citizens’ rights. The Court emphasised the importance of algorithmic transparency in the public administration and observed that public access to the source code enables social oversight, helps detect potential biases or errors, and builds trust in public digital tools. It held that exceptions to such access must be interpreted restrictively. Any entity denying access must identify the precise legal basis for the exception and provide a detailed, evidence-based justification showing a real, probable, and specific harm that outweighs the public interest in disclosure. Notably, the Court clarified that copyright does not constitute a valid ground to deny access to public information held by public entities. In reaching its conclusion, the Court referenced international standards and comparative practices, noting that many jurisdictions published the source code of similar COVID-19 tracking applications to promote transparency and public confidence.

The decision reinforces the principle of transparency in the use of AI tools and algorithms within the public administration, which will likely be relevant to the use of AI in the criminal context.