Brazil is at a crossroads. The new Brazilian Artificial Intelligence Plan (PBIA) 2024-2028 is a clear, ambitious vision to use AI to grow Brazil’s economy, improve its citizens daily lives, increase its competitiveness, and advance Brazil’s technological leadership on the world stage.
The global technology industry strongly supports Brazil’s ambition and agrees policymakers can and should leverage AI to solve immense challenges and unlock new opportunities for all Brazilians. To turn this vision into reality, the AI Bill (PL 2338/2023) will be a critical vehicle. Its goal of building trust is one tech shares and is the foundation for successful AI adoption in Brazil and worldwide. Our industry is committed to working with Brazil’s policymakers to preserve that fundamental value while allowing for continued innovation.
To that end, as policymakers work to refine this landmark legislation, we offer four key recommendations to ensure the bill strikes a balance between protecting individual rights and achieving Brazil’s ambition to be a leader in AI:
1. Calibrate Risk for True Proportionality
Think of a high-stakes football match. A good referee doesn't apply the same penalty to every foul. A light shirt-tug in midfield isn't treated the same as a dangerous, two-footed tackle in the penalty box. Applying a "one-size-fits-all" red card for every infraction would ruin the game and penalize skillful play.
AI regulation should function with the same proportionality and avoid creating duplicative regulatory regimes in areas where existing mechanisms already sufficiently address risk. Ensuring proportionality is critical, which is why we encourage policymakers in Brazil and globally to take a risk-based approach to advance the responsible development and deployment of AI. While we appreciate that Brazil has attempted to advance this approach by listing specific high-risk use cases in the current AI Bill, additional nuance is necessary to ensure that only truly high-risk applications – the “dangerous tackles” – are captured. The Bill should not sweep in entire classes of deployments or sectors; instead, it should articulate clear use cases that present significant risk to individual rights and safety and avoid sweeping in applications that are already regulated under sector-specific laws.
2. Focus Regulation on the "Frontier"
The bill’s current definition of “general purpose AI system” is very broad. As drafted, it would capture common, low-risk tools like translation or transcription software. Additionally, the bill introduces the concept of “systemic risk,” also defined in a broad manner. When taken together, these definitions create a regime that sweeps in a wide swath of AI technology that may not be particularly risky. It is similar to trying to apply the same safety rules to a bicycle, a family car, and a 40-ton truck.
A more effective approach is to focus rules on "frontier AI models" – the "40-ton trucks" of AI that have truly high-impact, next-generation capabilities. We recommend introducing a clear, updatable definition of a frontier AI model, and tying obligations to instances where a model has the potential to present catastrophic risk. This avoids capturing simple tools and focuses scrutiny where it matters most.
3. Advance a Balanced Copyright Regime
As policymakers consider how to address emerging questions around data governance and AI training, especially in the context of copyright, Brazil should seek to establish a balanced regime that protects rightsholders while promoting innovation and the development of its digital economy. We believe that any approach must allow for the lawful use of publicly available data, including content available on the web, while also ensuring appropriate safeguards are in place to comply with relevant privacy or copyright law.
The bill’s current draft, by prohibiting text and data mining (TDM) for commercial purposes and imposing unworkable rules with retroactive remuneration, could inadvertently cut off AI models from Brazil’s rich public data. This would result in services less attuned to the specific Brazilian cultural context. By embracing clear exceptions for TDM and promoting voluntary licensing models, Brazil can protect intellectual property while ensuring that AI services effectively serve all Brazilians.
4. Build Smart, Workable, and Coordinated Governance
A law’s vision is only as good as its practical application. A few key refinements to the bill’s governance structure would ensure it is workable, effective, and avoids unintended gridlock. Specifically, the bill should ensure this new competent authority, the National System for AI Regulation and Governance (SIA), is built with financial and technical autonomy, a single point of contact for business, and harmonization to avoid conflicting rules. Further, requirements like Algorithmic Impact Assessments (AIAs) must be clear. The bill must provide clarity on which party in the AI value chain needs to conduct an assessment to reduce duplicity, and ensure reports are only provided to the authority upon request to protect privacy. Finally, the Bill must establish clear definitions to help eliminate confusion or uncertainty.
Path Forward
Brazil has the talent and energy to realize its AI ambitions. The moment is right to fine-tune Bill 2338/2023 and make sure it is a catalyst for innovation. The technology industry supports this goal and looks forward to continuing working with policymakers in Brazil to ensure that the Bill strikes the right balance.