Does the Brazilian federal government have AI governance? What the official data shows

In Nov 2025, 214 of 234 Brazilian federal agencies reported no institutional AI policy or strategy. What the SISP data shows and what companies can copy.

Updated 14/nov/2026 · 9 min read

The short answer

The Brazilian federal government is adopting AI faster than it is building the structure to govern it. That is what the SISP Self-Assessment shows, the annual and mandatory survey run by the Secretariat of Digital Government (SGD/MGI), answered in November 2025 by 234 Federal Executive Branch agencies. In one year, deployed AI solutions went from 117 to 182. In the same survey, 118 agencies said they had no defined governance body for AI projects, and 214 said they had no institutionalized policy or strategy.

It is the same pattern seen in companies: use arrives first, and the policy, the owner and the risk process arrive later. This guide gathers the numbers on the use of artificial intelligence in federal public agencies, explains what the government is doing in response and lists what a company can copy without waiting. The data belongs to the federal government and describes it; it is not a measure of the private sector. For the definition of the topic, see what AI governance is.

214 of 234

Federal Executive Branch agencies reported having no institutionalized AI policy or strategy, and 118 reported no defined governance body for AI projects, in the SISP Self-Assessment of November 2025. It is a government survey of itself, not an audit.

Cetic.br | NIC.br (autoria SGD/MGI), Panorama Setorial da Internet, ano 18, n. 1: Inteligência Artificial no Poder Executivo Federal: governança, capacitação e soberania digital, p. 13

What the government measured, and how

The SISP Self-Assessment is a self-declared instrument, filled in every year and mandatory for the agencies that belong to the Information Technology Resources Administration System (SISP), established by SGD/MGI Ordinance No. 4,339/2023. The 2025 edition was run in November and drew answers from 234 Federal Executive Branch agencies. It measures IT governance in general and, since 2024, has tracked AI-specific indicators: strategy, solutions in use, owners, ethics, training and barriers.

The results appear in two official places. The SGD/MGI report holds the questionnaire tables. The Sectoral Overview of the Internet of April 2026, written by SGD staff and published by Cetic.br, part of NIC.br, analyzes the numbers. There is also an interactive dashboard at the Brazilian Artificial Intelligence Observatory (OBIA), with cuts of the same survey and 232 valid responses. This guide uses the numbers from the report and the Overview, which add up to 234.

AI use in public agencies is moving fast

In 2025 the survey counted 182 deployed AI solutions in 58 agencies. A year earlier it was 117 solutions in 41 agencies. Besides the deployed ones, there were 144 projects as pilots and 341 in prospecting.

Indicator20242025
Deployed AI solutions117182
Agencies with deployed solutions4158

The most cited reason to invest was improving the speed of public service delivery (181 agencies), followed by lowering the cost of delivering services (114) and creating new services (101). It is the same argument that moves companies: efficiency gains. What changes is the cost of ignoring governance, which efficiency gains do not pay for.

Governance is lagging

Ethical governance indicators also rose in 2025, but from a small base.

Indicator (number of agencies)20242025
With institutional guidelines or policies on ethical AI governance841
With an ethics committee623
That build ethical considerations into the solution lifecycle1822
With specific measures on ethical AI behavior611

Even with that progress, in 2025 there were still 118 agencies with no defined governance body for AI projects and 214 with no institutionalized policy or strategy. The questionnaire report shows the same picture from another angle: a five-level maturity scale for AI strategy, answered by the 234 agencies.

85%

of the 234 agencies are in the first two levels of the AI strategy maturity scale: 114 at level 1, with no structured initiatives, and 85 at level 2, with exploration under way and no formal strategy. Only 11 reached level 4 and 1 reached level 5.

Secretaria de Governo Digital (SGD/MGI), Autodiagnóstico SISP 2025: relatório de estatísticas do questionário, p. 216

LevelWhat it describesAgenciesShare
1No structured initiatives; one-off actions by technical teams, with no guidelines or policy11448.7%
2Exploration under way, with the first use cases but no formal strategy8536.3%
3AI governance foundations built into data governance; pilot projects239.8%
4Validated systems in operation and continuously improving governance114.7%
5AI consolidated as a strategic element, with fully integrated governance10.4%

The numbers at the start of the guide and those in this scale come from different questions, so they do not match to the digit. They tell the same story: most agencies are still starting to structure the governance of a technology they already use.

What holds it back: money, people and infrastructure

Among the barriers to developing AI applications, the most cited was a lack of financial resources (187 agencies). Next came difficulty hiring or retaining people with technology skills (170) and insufficient digital infrastructure (165). Because buying an external solution is complex, the main route has been building in-house, with civil servants: 127 agencies named it as the main one.

For anyone governing AI in a company, the lesson is familiar: a shortage of qualified people weighs more than a shortage of tools. The government responded with training at scale. The Brazilian Artificial Intelligence Plan aims to train 115,000 federal civil servants by the end of 2026.

The government’s answer: the Ethical Impact Self-Assessment

To support agencies, SGD’s Artificial Intelligence Hub developed in 2025 the framework for the Ethical Impact Self-Assessment of AI in the Public Sector (AIE), in line with international references such as the OECD and UNESCO. The AIE starts from four risk categories: algorithmic bias and discrimination, technical and security risks, legal vulnerabilities, and social and reputational risks.

  • Classification. A 12-question questionnaire places the solution in one of four risk levels: low, medium, high or excessive. A solution of excessive risk gets a recommendation to reassess the project.
  • Depth. Depending on the level, the team answers a simplified, standard or complete version of the questionnaire, organized around six axes: human rights, transparency, human oversight, security, privacy and sustainability.
  • Result. A report with an ethical adherence score from 1 to 5, charts per section and up to 44 improvement recommendations for the project team.

The logic is that of any good AI policy: classify the risk before deciding, and deepen the analysis according to the risk. It is the same risk-based approach found in the NIST AI RMF and in ISO/IEC 42001.

What a company can learn from these numbers

A private company does not answer to SISP, but the same failure points appear in any organization adopting AI. Five of them map directly to the government’s data:

  • Have someone answerable. In 118 of the 234 agencies there was no defined governance body. In a company, name an executive owner and a lean committee, with fast decisions and a record (see who answers for what).
  • Write the policy. In 214 agencies there was no institutionalized policy or strategy. In a company, a one-page policy with data examples already changes behavior (see the guide to an AI acceptable use policy).
  • Classify risk before approving. The AIE separates four levels before deciding. In a company, classify tools and use cases as allowed, restricted or prohibited, and go deep only where risk is high (see AI risks in the workplace).
  • Know what is in use. The government counted 182 deployed solutions, and only counted them because it asked. In a company, the use nobody declares is what weighs most; see how to detect which AIs employees use and the AI inventory.
  • Train people. In 170 agencies the difficulty was hiring or retaining qualified people. In a company, short and recurring training, with an owner in each area, costs less than any tool.

To measure where your company stands on these five points, the free assessment at /diagnostico shows, in 12 questions (3 about your company and 9 on maturity), about 3 minutes, a maturity score compared with the target for your sector.

Where Tangerin AI fits

Tangerin AI is the independent AI governance platform that discovers, assesses, governs and verifies AI use across the entire company, from any vendor.

From a company’s point of view, it covers the gaps the government found in itself: the inventory of real use, the classification of each tool by exposure level, the policy with a notice to the employee and a recorded acknowledgment, and the evidence for the board and for the auditor. Tangerin AI is a private company, is not part of the government and does not replace the AIE, which is the official instrument for federal agencies. See the plans and the 14-day trial, no card required.

Limits of this data

  • It applies to Brazil only. The data and rules are from the Brazilian government. Other countries have their own rules and surveys, and they are not covered here.
  • It is self-declared. Filling it in is mandatory, but the answers come from the agencies themselves and are not audited.
  • It covers only the Federal Executive Branch. That is 234 SISP agencies. The numbers do not apply to states, municipalities, the Judiciary, the Legislature or companies.
  • What about the Judiciary? It has its own rule, CNJ Resolution 615/2025. See AI in the Brazilian Judiciary: what CNJ Resolution 615/2025 requires.
  • It is a snapshot of November 2025. The comparison with 2024 comes from the survey itself, but the situation changes with each cycle.
  • It measures governance maturity, not Shadow AI. The numbers say whether there is a policy and an owner, not how much AI civil servants use on their own.
  • Sources with close but not identical bases. The OBIA dashboard uses 232 valid responses and the Cetic.br report 234, so the same indicator can differ by one or two units between the two.

Frequently asked questions

Do Brazilian public agencies have AI governance?

Not much yet. In the Federal Executive Branch, in November 2025, 214 of 234 agencies said they had no institutionalized AI policy or strategy, and 118 had no defined governance body for AI projects. The picture is improving: agencies with institutional guidelines or policies on ethical AI governance went from 8 to 41 in one year. The data covers the Federal Executive Branch, not the whole public sector.

What is the Ethical Impact Self-Assessment (AIE)?

It is the framework the Secretariat of Digital Government developed in 2025 so that federal agencies can identify and treat the ethical risks of an AI solution before they cause harm. It maps four risk categories, classifies the solution into four levels (low, medium, high or excessive), is organized around six axes (human rights, transparency, human oversight, security, privacy and sustainability) and produces a report with an ethical adherence score from 1 to 5 and up to 44 recommendations.

Where do these numbers come from, and how far can they be trusted?

From the SISP Self-Assessment 2025, an annual and mandatory survey run by the Secretariat of Digital Government, answered in November 2025 by 234 Federal Executive Branch agencies, and from the Sectoral Overview of the Internet of April 2026, published by Cetic.br. Because the answers are self-declared, the numbers show what agencies say about themselves, not what an audit would find. They are good for the order of magnitude and the direction, and weaker for comparing two agencies.

Why does government AI adoption show up with different numbers in each survey?

Because the surveys measure different things. The SISP Self-Assessment is self-declared and counts deployed solutions: 58 of the 234 agencies (about 25%) had at least one. Cetic.br's TIC Governo Eletrônico survey is sample-based and measures use of AI: in the 2023 data, 49% of federal agencies were already using the technology. Before comparing numbers, compare the definition behind each one.

What can a private company copy from the government?

Five things: a defined owner, a written policy, classifying each solution by risk level before approving it, an inventory of what is in use, and training for people. The government is taking each of them to federal agencies; a company can start today, without waiting for any regulation.

How many federal government agencies use artificial intelligence?

In November 2025, 58 of the 234 agencies (about 25%) had at least one deployed AI solution, for a total of 182 solutions, plus 144 pilot projects and 341 in prospecting. In the TIC Governo Eletrônico survey, with 2023 data, 49% of federal agencies reported using AI. The gap comes from the definition: one counts deployed solutions, the other measures any use.

Sources

Found it useful? Share it

LinkedInX

Instagram does not open links from outside the app, so we copy it for you to paste in a story or bio.

Where does your company stand today?

The free assessment has 12 questions (3 about your company and 9 on maturity), about 3 minutes, with no signup. It returns your maturity level per dimension and a prioritized action plan, enough to know where to start without buying anything.