MarIA at Brazil's Supreme Federal Court: transparency and AI in digital justice

STF stands for Supremo Tribunal Federal, Brazil's Supreme Federal Court. It is the country's highest constitutional court and a central actor in Brazilian public life.

The MarIA initiative, described in the original article as a module for AI-assisted drafting and procedural support, represents a broader movement inside courts: using artificial intelligence to manage massive caseloads, summarize materials and assist legal writing.

For international readers, the project should be understood less as a single tool and more as a case study in public-sector AI. When a constitutional court adopts AI, the question is not only whether the tool is efficient. The question is whether the tool is legitimate.

Courts face real workload problems. AI can help classify cases, identify repetitive arguments, draft summaries and reduce time spent on administrative tasks. Used carefully, it can improve access to justice by making institutions faster and more organized.

But judicial efficiency has a limit: it cannot turn legal reasoning into an opaque pipeline. Parties must understand the basis of decisions. Judges must remain responsible for the outcome. Public institutions must be able to explain how technological tools influence their work.

This is why due process is central. AI can support the court, but it cannot become an invisible decision-maker. Human review must be meaningful, not ceremonial.

Algorithmic transparency does not always mean publishing every line of code. In public-sector AI, it means that the institution can explain the tool's purpose, data sources, limitations, governance, audit mechanisms and role in the decision-making workflow.

The public should know whether AI is merely drafting a neutral summary, suggesting classifications, identifying precedents or influencing substantive outcomes. Each use carries a different risk profile.

Transparency is also institutional. Procurement, vendor relationships, source-code rights, model updates, logs and impact assessments matter because public power cannot be delegated into a black box.

The original article invokes international concerns about algorithmic discrimination, including the well-known COMPAS debate in the United States. COMPAS became a reference point because it showed how algorithmic tools in justice systems can raise fairness, bias and explainability concerns.

The lesson for Brazil is not that every judicial AI system will reproduce the same problem. The lesson is that courts must anticipate bias, test models, document limitations and avoid tools that parties cannot meaningfully challenge.

A democratic justice system must be able to tell citizens not only what it decided, but how technology participated in the path to that decision.

A responsible judicial AI program should include clear use cases, human supervision, audit logs, model-performance evaluation, bias testing, procurement transparency and channels for contesting errors.

It should also define prohibited uses. Not every task that can be automated should be automated, especially when liberty, fundamental rights or access to courts are at stake.

The safest approach is incremental: use AI first where it supports organization and drafting, then require stronger safeguards as the tool moves closer to substantive decision-making.

MarIA shows both the promise and the danger of AI in courts. The promise is a more efficient judiciary. The danger is a less transparent one.

The correct legal question is not whether courts should use AI. They will. The question is whether they will use it with enough transparency, accountability and humility to preserve the legitimacy of justice.