We educate Brazilian leaders and connect Brazilian researchers with the people and ideas shaping the field.
Make advanced AI and its safety questions understandable, so leaders can ask better questions and make informed decisions.
Understand advanced AI, assess its risks, and make informed decisions.
Proposed: Executive briefings · Decision workshops
Connect Brazilian researchers with frontier work, methods, and open questions—as learners and active contributors.
Connect Brazilian expertise with frontier methods, collaborators, and open questions.
Proposed: Technical discussions · Research exchange
Bring Brazilian and international perspectives into dialogue across research, policy, and institutional decisions.
Bring Brazilian and international perspectives together across research, policy, and practice.
Proposed: Cross-border roundtables · Shared questions
A shared question: how can human oversight remain effective as AI systems become more autonomous?
How can oversight constrain an autonomous coding assistant whose outputs cannot be fully trusted?
Which actions need approval, what must be tested before deployment, and who can pause the system?
Leaders bring operational constraints; researchers bring methods. Together, examine control protocols, their assumptions, and questions worth testing in Brazil.
An illustrative discussion grounded in coding-control research. Applying these approaches in other settings requires new evidence.
AI safety connects difficult scientific questions with decisions that affect people. Start with the essentials.
AI safety concerns understanding and reducing the risks of AI systems causing harm. It includes studying failures, preventing misuse, maintaining meaningful human oversight, and preparing for increasingly capable systems. A strong demonstration of capability is one piece of evidence; reliability and safety require their own scrutiny.
What can the system do? What evidence supports the safety claim, and what was not tested? How might it fail or be misused in this context? Who can intervene, and who is accountable? The answers should inform the limits, oversight, and review conditions for its use.
Open questions include how to evaluate dangerous capabilities, interpret model behavior, and keep increasingly autonomous systems under effective human control. A useful starting point is to identify a question, understand existing methods and their limits, and ask what evidence would meaningfully advance the answer.
Brazilian decisions benefit from contact with current scientific work, and international research benefits from Brazilian expertise, contexts, and perspectives. Two-way exchange can connect local questions with relevant methods, bring new people into the field, and create opportunities to contribute knowledge.
Find the questions closest to your work, and a useful place to begin.
The questionWhat do we need to understand before making a consequential AI decision?
Read leadership questionsThe questionWhere can Brazilian expertise contribute to open AI safety questions?
Explore research questionsThe questionHow can science, policy, and practice learn from one another?
See a shared questionLearning and exchange formats are proposed. The questions and reading on this page are available to explore now.