# The instruments States use to explain their algorithms

> The GPAI report analyzes the instruments States use to explain and make public sector algorithms transparent
> By Alfredo Collosa · Artificial Intelligence · 2026-01-20

The new typology of algorithmic transparency instruments is the central focus of the latest report published by the Global Partnership on Artificial Intelligence (GPAI) on the use of algorithms in the public sector. Under the [title](https://wp.oecd.ai/app/uploads/2025/06/2024-ATPS-A-state-of-the-art-report-of-algorithmic-transparency-instruments-ES.pdf) "Algorithmic Transparency in the Public Sector: A Report on the State of the Art of Algorithmic Transparency Instruments," the document analyzes how governments are trying to comply with transparency principles in the use of automated systems and what the main obstacles they face in that process are.

The report starts from the premise that algorithmic transparency is neither a single concept nor an isolated instrument, but rather a set of diverse mechanisms that can and should be analyzed systematically. In this context, the document's most relevant contribution is its proposal for a new typology of algorithmic transparency instruments, which makes it possible to organize existing practices and assess their real scope in democratic and accountability terms.

## Algorithmic transparency

Before developing this typology, the report establishes a broad conceptual framework for what algorithmic transparency should be understood to mean. Far from a simplistic definition, the document presents it as a multifaceted concept that can be interpreted simultaneously as a capacity, a principle, a norm, a right and an instrument.

First, it is defined as a capacity, insofar as the possibility of auditing, supervising and evaluating algorithmic systems. Also as a principle, which ethically guides the disclosure of relevant information. As a norm or rule, expressed in specific standards of accessibility and explainability. As a right, understood as the citizen's prerogative to know how the algorithms that affect their life work; and as an instrument, materialized in specific mechanisms that make that transparency effective.

The report links this notion with other fundamental principles of digital public governance, such as explainability, accountability and the right of access to information. In operational terms, it defines algorithmic transparency as "the capacity of actors internal or external to the development of an algorithm to obtain information, supervise, test, criticize or evaluate the logic, procedures and performance of an algorithmic system in order to foster trust and increase the responsibility of the system's developers or controllers."

## The typology, the report's guiding thread

On this conceptual basis, the report introduces its main contribution: a typology that distinguishes between supply-driven algorithmic transparency instruments and demand-driven instruments. This distinction structures the entire subsequent analysis and serves as the document's guiding thread.

**[Alfredo Collosa publishes a book on AI in the public sector](https://observatorioblockchain.com/ia/alfredo-collosa-publica-un-libro-sobre-la-ia-en-el-sector-publico/)**

Supply-driven instruments are characterized by the proactive disclosure of information by the State, without the need for a prior request. These include public algorithm repositories, periodic publication of information on automated systems, automated responses integrated into administrative procedures and the disclosure of algorithmic audits.

By contrast, demand-driven instruments are activated in response to citizen requests, journalistic investigations or administrative and judicial requirements. This group includes requests for access to public information and disclosures ordered by competent authorities.

According to the report, this typology not only makes it possible to classify existing instruments, but also to assess the degree of maturity of public policies on algorithmic transparency and to detect imbalances between reactive and proactive approaches.

## Public algorithm repositories

Among supply-driven instruments, the report pays special attention to public algorithm repositories, considered one of the most promising mechanisms for guaranteeing citizen access to relevant information on automated decisions in the public sector.

One of the document's main empirical contributions is the mapping of 83 public algorithm repositories, active or under construction, distributed across different levels of government and regions of the world. Of these, 90% are active. The United States clearly leads with 37 repositories, followed by the Netherlands, France and Colombia.

However, the report stresses that the adoption of this type of instrument is uneven. Only 11 of the 29 GPAI member countries have at least one active repository, which highlights significant gaps in the implementation of algorithmic transparency policies.

Most of the repositories have been developed by national bodies, although subnational, academic and civil society initiatives are also identified. As for their content, they usually include basic information such as the system's name, its objectives, the responsible unit, the technical architecture, the data sources used and the handling of personal data.

Nevertheless, the report identifies relevant shortcomings: the disclosure of performance metrics, social impacts, review mechanisms or discrimination risks is limited or nonexistent in many cases.

## Auxiliary instruments and citizen understanding

Some more advanced repositories incorporate auxiliary instruments that seek to improve citizen understanding and facilitate interaction with algorithmic systems. These elements include glossaries, explanatory manuals, feedback forms, personalized alerts, interactive simulators and social media outreach strategies.

According to the report, these additional components are key to preventing algorithmic transparency from being limited to a mere accumulation of technical information that is difficult to interpret, and to moving toward more accessible and participatory models.

## Current limits

From a democratic perspective, the report identifies eight fundamental reasons that justify the need to develop robust algorithmic transparency instruments. These include strengthening the right of access to information, protecting fundamental rights, promoting an open State, applying ethical principles in artificial intelligence, public oversight of automated decisions, improving institutional efficiency, organizational learning and generating useful information for the technology ecosystem.

However, the document warns that many of the current instruments, including repositories, have significant limitations. In numerous cases, the information published is superficial, fragmentary or difficult to interpret, which can lead to "performative transparency" with no real effect on accountability.

Among the main challenges are the lack of institutional capacities, the technical and financial costs of maintenance, restrictions arising from intellectual property or cybersecurity, and the absence of common standards. Added to this is the fact that information is often available in only one language, which limits its accessibility in multilingual contexts.

## Conclusions

The report concludes that algorithmic transparency should not be reduced to the simple publication of data, but should make it possible to evaluate, challenge and improve the automated systems used by the public sector. In this sense, it proposes moving toward minimum disclosure standards, the creation of centralized and multilingual repositories, the integration of citizen participation mechanisms and the incorporation of explainable artificial intelligence technologies as a complement to existing transparency instruments.

It also recommends the development of single portals per country or region, with homogeneous publication criteria, that facilitate access to, comparison of and oversight of public algorithmic systems.

Although it is presumed that these instruments can increase public trust, improve administrative efficiency and reduce information requests, the report acknowledges that there is not yet enough empirical evidence to confirm this. It therefore proposes as a priority line of research to evaluate whether algorithmic transparency instruments, as currently designed, really generate public value.
