AI Governance in Practice
Real-world examples of how artificial intelligence impacts critical sectors in Brazil — and why responsible governance is indispensable in each of them.
AI Governance is not an abstract concept. Artificial intelligence systems already make or influence decisions that affect the lives of millions of Brazilians — from granting social benefits to medical diagnoses, credit decisions and court rulings. Without adequate governance, these systems operate as black boxes with real power over fundamental rights.
On this page, we present 15 concrete use cases across 5 sectors of Brazilian society, showing the specific governance challenges in each context and IBGIA's proposals to address them. To explore concepts further, see our AI governance glossary and the page on the AI governance framework in Brazil.
Public Sector
AI in decisions that affect citizens' fundamental rights
AI systems in courts
The Brazilian judiciary already uses more than 63 AI systems, including sentence prediction tools, automatic case screening and document analysis. Systems such as Victor (STF) and Elis (TJMG) process thousands of cases, but raise serious questions about algorithmic bias and lack of transparency in automated decisions.
Governance Challenges
- Algorithmic bias in judicial predictions that may perpetuate historical inequalities
- Lack of transparency about how algorithms influence procedural decisions
- Absence of independent audit mechanisms for the systems used
- Risk of eroding the principle of the natural judge and the duty to provide reasoning for decisions
What IBGIA proposes
- Implement mandatory algorithmic audit frameworks for judicial systems, with periodic publication of bias and performance reports
- Establish explainability requirements for any AI system used in decisions affecting fundamental rights, in line with PL 2338/2023 principles
Facial recognition in public safety
The use of facial recognition by law enforcement has been expanding rapidly in Brazil, with systems deployed in airports, stadiums and public roads. International studies and Brazilian cases show significantly higher false positive rates in Black populations, leading to wrongful arrests and rights violations.
Governance Challenges
- Disproportionate error rates in Black populations — studies show up to 10x more false positives
- Absence of a specific legal framework for the use of facial biometrics by public authorities
- Lack of impact studies before large-scale deployment
- Chilling effect on the rights of assembly and protest
What IBGIA proposes
- Require algorithmic impact assessments before any facial recognition deployment, including racial and gender bias analysis
- Create a moratorium on use in public spaces until minimum accuracy and equity standards are established by regulation
Algorithms in social benefits
Algorithms are increasingly used to determine the granting of benefits such as INSS, Bolsa Familia and BPC. Automated systems analyze beneficiary profiles, calculate eligibility and even detect fraud — decisions that directly impact the survival of millions of vulnerable Brazilians.
Governance Challenges
- Automated decisions on social rights without effective possibility of human review
- Training data that may reflect historical discrimination against vulnerable populations
- Lack of transparency for beneficiaries about algorithmic eligibility criteria
- Risk of digital exclusion: populations most in need of benefits have the least access to technology
What IBGIA proposes
- Guarantee the right to human review for all automated decisions that deny or revoke social benefits
- Implement accessible explanation mechanisms so beneficiaries understand how and why decisions were made
Healthcare
AI transforming diagnostics, triage and clinical research
AI-assisted diagnosis
AI systems for cancer detection, medical imaging analysis and diagnostic support are being adopted by Brazilian hospitals and laboratories. These tools can significantly increase diagnostic accuracy and speed, but raise complex questions about ANVISA regulatory approval and medical liability.
Governance Challenges
- Regulatory approval: ANVISA is still developing frameworks for AI-based Software as a Medical Device (SaMD)
- Medical liability: who is responsible for an AI-assisted misdiagnosis — the physician, the hospital or the developer?
- Bias in training data: models trained primarily on Northern Hemisphere populations may underperform for the Brazilian population
- Lack of large-scale clinical validation with Brazilian population data
What IBGIA proposes
- Develop specific guidelines for validation and certification of AI systems in healthcare, in collaboration with ANVISA, CFM and medical societies
- Require AI diagnostic systems to include information about the populations on which they were trained and validated, with mandatory testing on the Brazilian population
Patient triage in the public health system (SUS)
Algorithms are being used to prioritize care in the Unified Health System (SUS), from risk classification in emergency rooms to queue ordering for surgeries and specialized consultations. In a system with limited resources, these algorithmic decisions literally determine who is seen first.
Governance Challenges
- Algorithmic prioritization criteria that may reproduce existing access inequalities in the SUS
- Lack of transparency for patients about how AI influences their position in the queue
- Risk of underreporting emergencies in populations with lower health literacy
- Ethical questions about delegating scarce resource allocation decisions to algorithms
What IBGIA proposes
- Create hospital AI ethics committees to oversee implementation and continuously monitor the outcomes of algorithmic triage systems
- Ensure patients have access to clear information about how AI participates in decisions about their care, with the right to contest
AI in clinical research
Artificial intelligence is accelerating treatment development, from the discovery of new drugs to the analysis of large volumes of clinical data. In Brazil, institutions such as Fiocruz and reference hospitals already use AI for epidemiological data analysis and pattern identification in electronic health records.
Governance Challenges
- Patient data governance: ensuring that sensitive health data is used ethically and in compliance with the LGPD
- Informed consent: patients do not always know that their data will be used to train AI models
- Risk of concentration: technology companies with greater data access may monopolize AI-based research
- Scientific reproducibility: the opacity of some AI models makes peer validation difficult
What IBGIA proposes
- Establish health data governance protocols for AI research, including anonymization, consent and responsible sharing requirements
- Promote the creation of Brazilian health data repositories for AI research, with public governance and regulated access
Financial
AI in credit decisions, fraud detection and Open Finance
Algorithmic credit scoring
AI models determine credit access for millions of Brazilians, analyzing hundreds of variables to calculate risk scores. Beyond traditional credit bureau data, new models incorporate alternative data such as social media behavior, mobile usage patterns and geolocation — expanding the risks of discrimination.
Governance Challenges
- Algorithmic discrimination: models may use proxy variables for race, gender or location, violating the LGPD and the Consumer Protection Code
- Opacity: consumers struggle to understand why they were denied credit and how to improve their score
- Use of alternative data without explicit consumer consent or understanding
- Cumulative effects: a low score can impact access to housing, employment and other essential services
What IBGIA proposes
- Require periodic bias assessments in scoring models, with publication of equity metrics by demographic groups, as recommended by BACEN
- Regulate the right to explanation for automated credit decisions, ensuring consumers understand the determining factors and can contest them
Fraud detection and anti-money laundering
Financial institutions use AI to detect suspicious transactions, identify money laundering patterns and prevent fraud in real time. These systems process billions of transactions, but must balance detection effectiveness with privacy protection and minimizing false positives that block legitimate customers.
Governance Challenges
- Balance between security and privacy: excessive monitoring may violate customer rights
- False positives that improperly block transactions from legitimate customers, with disproportionate impact on small businesses and lower-income populations
- Opacity of detection criteria: blocked customers often do not receive adequate explanation
- Constant evolution of fraud techniques requires continuous model updates, with risks of algorithmic drift
What IBGIA proposes
- Create governance frameworks for anti-fraud systems that establish clear monitoring limits and guarantee contestation rights
- Promote a regulatory sandbox with BACEN to test new AI approaches to fraud detection with oversight and social impact metrics
AI in Open Finance
With the advancement of Open Finance in Brazil, AI systems now analyze shared banking data across institutions to offer personalized products, financial profile analyses and automated recommendations. The governance of shared data and the transparency about how AI uses it are central challenges.
Governance Challenges
- Governance of shared sensitive data: who is responsible when data flows between multiple institutions?
- Granular consent: consumers often do not understand the extent of sharing and use of their data
- Risk of concentration: large technology platforms may accumulate disproportionate informational advantages
- Cybersecurity: attack surface expanded by data sharing across multiple agents
What IBGIA proposes
- Develop AI governance standards specific to the Open Finance ecosystem, in partnership with BACEN and the financial sector
- Require algorithmic transparency in all automated financial recommendations, with clear indication that an AI participated in the decision
Legal
AI in case analysis, contracts and regulatory compliance
Jurimetrics and predictive analysis
Jurimetrics — the application of quantitative methods and AI to Law — has been growing significantly in Brazil. Predictive analysis tools are used by the CNJ, superior courts and large law firms to estimate probabilities of success in cases, identify decision-making patterns and optimize legal strategies.
Governance Challenges
- Risk of self-fulfilling prophecy: if lawyers adjust strategies based on AI predictions, the predictions shape the outcomes themselves
- Bias in historical data: past decisions reflect inequalities and may not represent desirable jurisprudence
- Lack of standardization in the collection and structuring of judicial data across Brazilian courts
- Limited transparency about methodology and limitations of commercial jurimetrics tools
What IBGIA proposes
- Establish guidelines for the ethical use of jurimetrics, including methodological transparency requirements and warnings about prediction limitations
- Promote the creation of open and standardized judicial datasets, facilitating independent research and auditing of commercial tools
Smart contracts and automated review
AI tools for contract review, analysis and generation are becoming common in legal departments and law firms. These systems can analyze thousands of contracts quickly, identify risk clauses and even generate complete drafts — but raise questions about legal liability and reliability.
Governance Challenges
- Legal liability: who is responsible for errors in AI-generated or reviewed contracts — the lawyer, the firm or the technology provider?
- Hallucinations in generative AI: models may invent nonexistent clauses, case law or regulatory references
- Confidentiality: sensitive contractual data may be used to train models, creating leakage risks
- Replacement of human analysis: pressure for efficiency may lead to uncritical acceptance of AI outputs
What IBGIA proposes
- Develop quality and certification standards for legal AI tools, with mandatory human validation requirements
- Create guidelines on the lawyer's professional liability in using AI tools, in collaboration with the OAB and the CNJ
Automated compliance
AI systems are being used to monitor regulatory compliance in real time, track legislative changes, identify non-compliance risks and generate automatic reports. With the growing complexity of the Brazilian regulatory environment — LGPD, PL 2338/2023, sector-specific regulations — compliance automation is becoming increasingly relevant.
Governance Challenges
- Regulatory complexity: systems must track multiple evolving regulations simultaneously (LGPD, PL 2338, sector-specific regulations)
- Normative interpretation: AI may not capture interpretive nuances and jurisprudential divergences
- Risk of false security: organizations may treat automated compliance as a complete substitute for human legal analysis
- Accountability: difficulty in tracing how the system reached a compliance or non-compliance conclusion
What IBGIA proposes
- Promote the integration of automated compliance tools with AI governance frameworks, ensuring human oversight in critical decisions
- Develop continuous validation methodologies for AI-based compliance systems, with regular testing against real-world scenarios and edge cases
Education
AI in personalized learning, assessment and school management
Adaptive learning systems
AI-personalized learning platforms adapt content, pace and methodology to each student's profile. In Brazil, these tools are being adopted in both public and private schools. However, data collected about children and adolescents requires special protection and rigorous governance.
Governance Challenges
- Protection of children's and adolescents' data: the LGPD requires parental consent and treatment based on the best interests of the child
- Commercialization of educational data: private platforms may use learning data for commercial purposes without transparency
- Excessive standardization: algorithms may restrict pedagogical diversity by optimizing for narrow performance metrics
- Digital exclusion: students without adequate access to technology fall increasingly behind as education becomes digitized
What IBGIA proposes
- Create national guidelines for the ethical use of AI in education, with special focus on the protection of minors' data and transparency for parents and guardians
- Require impact assessments before adoption of AI platforms by public school systems, including equity and accessibility analysis
Automated assessment
AI tools for grading exams, evaluating essays and analyzing academic work are increasingly used in Brazil — from the ENEM to internal university assessments. These tools promise efficiency and consistency, but raise concerns about linguistic bias, equity and the ability to evaluate critical thinking.
Governance Challenges
- Linguistic bias: systems trained on the writing patterns of certain groups may penalize regional and sociolinguistic variations
- Equity: students with access to generative AI for writing may benefit in automated assessments that do not detect AI use
- Qualitative limitation: AI struggles to evaluate creativity, originality and critical thinking
- Pressure on teachers: reducing assessment to quantifiable metrics may devalue human pedagogical judgment
What IBGIA proposes
- Develop quality and equity standards for automated assessment tools, with mandatory testing for linguistic and socioeconomic bias
- Ensure that automated assessment is always complementary — never a substitute — to human assessment in high-impact decisions (pass/fail, selection)
Algorithmic school management
Algorithms are used in school management for resource allocation, dropout prediction, student distribution across classes and demand planning. Education departments in large Brazilian cities already use predictive models to identify students at risk of dropping out.
Governance Challenges
- Transparency for parents and students: algorithmic decisions about school allocation and resources are often invisible to the school community
- Stigmatization: students identified as 'at risk' by algorithms may suffer negative effects from premature labeling
- Data quality: Brazilian educational data is frequently incomplete, outdated and inconsistent across school systems
- Reduction of the human role: excessive reliance on data may distance administrators from direct contact with school realities
What IBGIA proposes
- Implement transparency mechanisms that allow the school community to understand how algorithms participate in decisions affecting students and schools
- Establish protocols ensuring human oversight and pedagogical contextualization of all algorithmic recommendations in school management
Receive practical AI governance cases
Subscribe to the IBGIA newsletter and receive sector analyses, frameworks and case studies on responsible AI.
Exclusive AI governance insights
Get weekly analyses, regulatory alerts and practical frameworks on AI governance in Brazil — free.
Deepen your knowledge
Want to help build AI governance in Brazil?
Join IBGIA and contribute your experience to tackle the practical challenges of AI in every sector of society.