EU AI Act
We apply risk-based classification principles from the EU Artificial Intelligence Act (Regulation 2024/1689), treating high-risk AI applications with heightened scrutiny, human oversight and documentation requirements.
EU AI Act · NIST AI RMF · OECD AI · UNESCO · LGPD
AI POLICY
Artificial intelligence with judgment, context and accountability.
AI is not a shortcut, an absolute promise or a substitute for technical responsibility. At 33 Digital, it is a support layer — used to amplify human capacity, never to mask uncertainty or replace qualified judgment. This policy is addressed to partners, clients, institutional bodies and international regulators.
Responsible useClear limitsData governanceHuman oversight
REGULATORY ALIGNMENT
Our AI practices are aligned with leading international frameworks.
33 Digital is incorporated in Brazil and operates under the LGPD. We voluntarily align our AI practices with international standards to ensure consistent, accountable and transparent use of artificial intelligence across all our products and services.
NIST AI RMF
Our AI development and deployment practices are informed by the NIST AI Risk Management Framework (AI RMF 1.0), covering Govern, Map, Measure and Manage functions for responsible AI lifecycle management.
OECD AI Principles
We adhere to the OECD Principles on AI (2019, updated 2024): inclusive growth, human-centred values, transparency, robustness and accountability — applied across all AI-powered tools we build.
UNESCO Recommendation
We align with the UNESCO Recommendation on the Ethics of AI (2021), particularly on human rights, non-discrimination, fairness, sustainability and the protection of vulnerable groups.
Brazilian regulation
Our primary applicable law is the LGPD (Law No. 13,709/2018). We also monitor the Brazilian AI Bill (PL 2338/2023) and ANPD guidance on automated decision-making as the regulatory landscape evolves.
Last updated
January 2026. This policy is reviewed periodically as our practices evolve and international AI regulation matures. Material changes will be communicated on this page.
HOW WE USE AI
AI as a support layer, not a final answer.
We use artificial intelligence to organize information, interpret data, automate processes and expand analytical capacity. Every application starts from a real problem, with a declared purpose and human review where required.
Data structuring
Organizing public or authorized data, documents, workflows and information to make them useful, traceable and interpretable by people and systems.
Diagnosis and triage
Supporting initial analyses, automated triage and pattern identification in operational contexts with defined human supervision at critical decision points.
Agents and automation
Building service agents, automated workflows and assistants connected to real data, business rules and specific operational context — with traceable logic.
Research and development
Accelerating research, context organization and proprietary tool development with AI as an accelerator — not as a substitute for method, expertise or human judgment.
We do: use AI to expand access to information, organize context, automate repetitive processes and support decisions with real data and declared purpose.
We do not: use AI to replace accountability, mask uncertainty, create an illusion of precision or operate without the possibility of human review and contestation.
PROHIBITED USES
What AI must not do at 33 Digital — unconditionally.
Defining what we do not do is as important as declaring what we do. These prohibitions are not operational constraints — they are ethical commitments that guide every architecture and deployment decision, regardless of commercial pressure or technical convenience.
No replacement of human judgment
AI does not make final decisions in sensitive contexts — legal, social, financial, health-related or security-related. These contexts require qualified human assessment. (Aligned with EU AI Act Art. 14 — human oversight.)
No manipulation or exploitation
AI must not be used to exploit vulnerabilities, manipulate behavior, create intentional biases or produce conclusions without context and without the possibility of contestation. (Aligned with EU AI Act Art. 5 — prohibited practices.)
No unsupervised operation in critical contexts
Automated outputs in high-impact contexts require human review, technical validation and a clearly declared purpose before any use. Opacity in decision-making is not acceptable.
No deceptive content generation
We do not use AI to produce disinformation, simulate identities, fabricate evidence or generate content that leads people to harmful decisions. Transparency about AI-generated content is mandatory.
High-risk AI classification
Applications involving critical infrastructure, access to essential services, law enforcement, migration, justice or fundamental rights are treated as high-risk under EU AI Act criteria and subject to the strictest internal controls.
Auditability requirement
AI systems built by 33 Digital are designed to allow decision traceability, error identification and correction — without relying on opaque black-box processes that cannot be explained or challenged.
AI to expand capacity. Never to mask uncertainty.
DATA, PRIVACY AND AI GOVERNANCE
Public data is not free territory. Personal data demands even greater care.
Every AI application involving data — public or private — is handled with responsibility, declared purpose and respect for privacy. The origin of data does not eliminate the ethical obligation to use it with judgment, security and proportionality.
Data minimization
We collect and process only the data necessary for the declared purpose. Unnecessary data is not retained, cross-referenced or reused without a new lawful basis. (LGPD Art. 6 · GDPR Art. 5(1)(c).)
Purpose limitation
Each AI use case involving data has a specific, known and documented purpose. Data is not repurposed for undeclared uses after collection. (LGPD Art. 6 · GDPR Art. 5(1)(b).)
Transparency about automation
When a response, triage or decision is generated or supported by AI, this is identifiable. We do not conceal the use of automation in interactions that may affect people. (EU AI Act Art. 50 · LGPD Art. 20.)
Non-discrimination
AI systems must not produce outputs that discriminate based on race, ethnicity, gender, religion, disability, sexual orientation or any other protected characteristic. Bias monitoring is part of our development process.
Automated decisions: individuals have the right to request human review of any significant decision made solely by automated means, in accordance with LGPD Art. 20 and GDPR Art. 22.
AI incident response: in the event of an AI-related incident causing harm or risk to individuals, 33 Digital will take corrective action and notify affected parties in accordance with applicable law.
COMMITMENT
Our commitment is to use AI to expand capacity, not to replace accountability.
When a solution requires human review, technical validation, a trusted source or additional care, that threshold will be preserved — regardless of commercial pressure, deadline or technical convenience. This policy is part of the architecture, not a footnote. Institutional partners, regulators and data subjects are welcome to contact us with questions or concerns.
Judgment
Context
Transparency
Limits
Human oversight
AI IN SOCIAL IMPACT TOOLS
When AI operates in sensitive contexts, the constraints are even stricter.
Tools such as Quilombo Digital and Mulher Digital involve community data, vulnerability histories and protection contexts. In these cases, AI operates under additional restrictions, mandatory human oversight and reinforced privacy controls.
Guidance, not decision
AI in social tools organizes information, presents pathways and supports context. It never decides for people in situations of vulnerability, risk or protection — human judgment is always preserved.
Reinforced privacy
Data from quilombola communities, women in risk situations or vulnerable groups receives additional protection — restricted access, encryption and strictly declared purpose, aligned with LGPD sensitive data provisions.
No substitution of official networks
No 33 Digital tool replaces police, public defenders, social services, health systems or any official protection and support network. This is a non-negotiable design constraint.
Responsible referral
In situations of risk, urgency or emergency, tools direct users to official channels. No automated flow retains a person when the situation requires immediate human intervention.
Vulnerable groups protection
Consistent with UNESCO Recommendation on AI Ethics (2021) and EU AI Act provisions on prohibited AI practices, we apply the highest level of care when AI interacts with or affects vulnerable individuals or communities.
Emergency contacts (Brazil)
In emergencies: 190 (Military Police), 192 (SAMU — Emergency Medical Services), 180 (Women's Support Hotline). No digital tool replaces emergency response.