Why AI regulation could kill startups in Brazil

The regulation of artificial intelligence (AI) has gained worldwide attention, with countries seeking to establish clear guidelines for the development and application of this technology. In Brazil, this discussion is advancing with the processing of Bill No. 21/2020 in the Federal Senate. As a technology entrepreneur and startup specialist, I closely follow these debates, recognizing the importance of regulation that promotes innovation without imposing excessive barriers to technological development.

The aforementioned bill aims to establish foundations, principles, and guidelines for AI in the country. While the initiative seeks to create a regulatory framework, it is crucial to analyze how its specific provisions may impact the startup ecosystem and technological innovation in Brazil.

Specific Points of PL 21/2020 and Its Impacts:

• Fundamentals and Principles:
  • The Bill establishes principles such as human-centeredness, non-discrimination, transparency and explainability. Although fundamental, the lack of clarity about its practical application can generate uncertainty for developers and entrepreneurs.
• Broad Definitions of AI Agents:
  • The bill defines “AI agents” as those involved in the development and operation of AI systems, without clearly specifying the responsibilities of each party. This generalization could result in legal uncertainty, especially for startups that collaborate with third parties in the development of AI solutions.
• Unclear Definition of “High Risk”:
  • The Bill mentions the need for governance measures for AI systems classified as “high risk”. However, it does not clearly specify the criteria that determine this classification, creating uncertainty for developers and entrepreneurs when trying to identify the regulatory obligations applicable to their systems.
• Data Mapping and Audit Requirements:
  • For AI systems, especially high-risk ones, the project suggests carrying out impact analyses and mapping the risks associated with the expected uses of the system, as well as possible forms of misuse. In addition, it proposes governance of the data used for training, testing and validation of the system, as well as continuous control and auditing. These requirements may be unfeasible for startups due to the costs and resources required to implement complex compliance, documentation and auditing processes. In some cases, depending on the architecture of the AI model, it may be technically impossible to map in detail all the data used, especially when using deep learning models that operate as “black boxes”.
• Artificial Intelligence Impact Report (AIIR):
  • Similar to the LGPD’s Data Protection Impact Report, the bill proposes the RIIA to document the life cycle of AI systems and risk mitigation measures. Preparing such a report can be costly and complex for startups with limited resources.
• Sanctions and Penalties:
  • The bill provides for sanctions ranging from warnings to significant fines and suspension of AI-related activities in case of non-compliance. For startups, these penalties can be disproportionate and threaten the viability of the business.
• Indication of Additional Regulations:
  • The text of the bill indicates that additional regulations still need to be constructed to detail certain aspects. This need for future regulations creates uncertainty for startups and developers, who may face difficulties in planning and developing their AI systems without clear and established guidelines.

Implications for Startups and Innovation:

The lack of clarity and detail in key points of Bill 21/2020 may result in an uncertain regulatory environment, discouraging startups and innovation in the AI sector. Emerging companies may face challenges in interpreting and complying with vague principles, in addition to dealing with possible poorly defined legal responsibilities. For AI regulation in Brazil to be effective and promote innovation, it is essential that the bill be enhanced with clear and applicable guidelines, developed in dialogue with the technology ecosystem and civil society.