Human autonomy
AI should expand people’s choices and capabilities without taking away information, agency, or the opportunity for review.
Responsible AI
Technology with human direction
Intelligence expands.
Humanity guides.
We use artificial intelligence to expand human capability — not to replace responsibility, listening, or qualified judgment. We define purposes and limits, assess quality and risk, protect data, and maintain human oversight throughout the life cycle.
PurposeOversightTransparencyAccountability
01 / A framework for direction
Six principles to guide what we build.
These principles set our institutional direction. Their practical application must be documented for each product, taking into account its use, audience, data, risks, and specific consequences.
Well-being and safety
Expected benefits, risks, misuse, and consequences must be assessed in proportion to the context.
Meaningful transparency
People should know when AI is involved, what purpose it serves, and which limitations matter to their decision.
Human accountability
Roles, oversight, avenues for challenge, and accountability must remain clear throughout the entire life cycle.
Inclusion and equity
Data, models, and experiences should be assessed for barriers, bias, and unequal impacts across people and groups.
Responsible evolution
Quality, safety, context, and unexpected effects must be continuously monitored, corrected, and documented.
02 / Continuous governance
Responsibility before, during, and after.
Define purpose and limits
- user and context
- intended use
- data required
- risks and oversight
Evaluate and document
- quality and performance
- errors and bias
- security and privacy
- appropriate explanations
Monitor and respond
- real-world effects
- changes in context
- incidents and corrections
- suspension when necessary
03 / Generative AI
Fluency is no guarantee of truth.
Generative models can produce plausible but false answers, omit context, reproduce bias, or invite excessive trust. In health and other consequential decisions, content requires appropriate human verification before it informs any action.
incorrect answersover-automationbiasprompt privacyincomplete context
04 / Personal data
Privacy by design.
Data use must follow a legitimate purpose, necessity, security, and the rights established by applicable law. Sensitive data, including health data, requires enhanced protection.
When a decision is made solely through automated processing and affects a person’s interests, the rights to review and information provided under the LGPD must be considered.
Institutional references
Official principles and guidance.
Building together