[Flash Report] Brazil's Artificial Intelligence Plan / AI for the Good of All and Law / Random Thoughts on Reading the Brazilian AI Plan PBIA

0 Introduction

Discussions on AI governance tend to be drawn toward the topic of "regulation." Is it tort law, product liability, personal data, or copyright? Of course, these are important. However, when a nation faces AI, it seems that what first emerges is that country's culture and narrative. For what purpose is AI used, whose interests are prioritized, and to what extent does the state accept responsibility? Brazil's Brazilian Artificial Intelligence Plan (PBIA), "AI for the Good of All," seems to be a document that is realistic in how it constructs this narrative.

1 "What to list in the name of good"

The PBIA positions AI as "AI for the Good of All" and presents the conditions for this as five pillars. The vocabulary listed there is interesting. Human-centricity, inclusion, respect for labor, SDGs, the right to development, national sovereignty, transparency, explainability, privacy, cybersecurity, intellectual property, and democracy and information integrity. Slogans for AI policy abound around the world, but when "rights," "sovereignty," and "democracy" are lined up on the same level like this, it is closer to a constitutional preamble than an ethical declaration.

This arrangement can also be seen as a provocation against the debate on AI legislation, depending on one's perspective. That is, it is not a question of what obligations to impose to curb AI risks, but rather a question of which values to "fix first" in the process of nurturing AI as a foundation for society. Values become mere decoration if added later, but become a blueprint if placed first. The PBIA appears to have chosen the latter.

2 Another set of fundamental rights: "The right to development" and sovereignty

Pillar III brings "the right to development" and "national sovereignty" to the forefront. It may sound a bit radical to the ears of legal scholars, but the reality surrounding AI is radical to begin with. Computing resources, data, semiconductors, cloud, standardization, and research talent—these are no longer just market goods, but resources that determine the range of motion for a nation's decision-making. The PBIA's mention of vigilance against "dependence on foreign solutions" and technological curtailment is also an extension of this.

Here, the role of law becomes twofold. First, law as a shield to prevent rights violations. Second, law as a lever to promote capacity building toward technological independence. The latter is often dismissed as so-called "industrial policy," but it is actually the core of governance. This is because whether a nation settles for importing AI or nurtures it as a public good changes its future regulatory capacity, auditing capacity, and even the effectiveness of its democracy.

3 Is regulation only for "stopping" things?

Pillar IV of the PBIA, while upholding transparency, traceability, and accountability, states that it is technically difficult to determine where responsibility lies in the AI value chain (development, provision, and use). This point is painfully honest for legal scholars. This is because responsibility theory has been designed (at least in principle) on the premise that the point of attribution is visible. However, with AI, the result changes if any one of the training data, model design, fine-tuning, operation, prompts, peripheral systems, or monitoring systems is missing. Law desires a "map" of causality and attribution, but AI makes that difficult to see.

Against this fog, the law can take two attitudes. First, an attitude of disliking the fog and strengthening prohibitions and chilling effects. Second, an attitude of injecting accountability and auditability into the system, assuming the fog exists. The PBIA leans toward the latter. At the very least, as long as it claims to be "AI for the Good of All," a declaration of determination that it cannot end with just prohibitions is visible.

4 The time design of a "two-stage approach"

The PBIA adopts a two-stage approach that combines immediate impact actions, which show results in the short term, with structuring actions, which build capacity in the medium to long term. Here, "time," which legal discussions tend to overlook, is brought in. In AI policy, time is not just a schedule. It is the collision point of the speed of technological competition, the speed of social implementation, and the speed of regulation and relief. The two-stage approach is a device to face that collision head-on.

Furthermore, the PBIA anticipates an investment of approximately 23 billion reais between 2024 and 2028. When numbers are included, policy turns from air into a structure. Conversely, AI principles without numbers usually end up as clouds.

5 Sovereign Cloud is an 'Infrastructure Law' Issue

The PBIA symbolically highlights the sovereign cloud. As a cloud infrastructure operated by public institutions or entities under public management, it is expected to ensure the protection of sensitive data, domestic storage, data sovereignty, and privacy. What should be noted here is that AI governance (even before considering the morality of models) is encroaching upon the governance of infrastructure. Data centers, computing resources, cataloging of government data, and interoperability—these have traditionally tended to be dismissed as matters of 'administrative information policy' or 'procurement.' However, in the AI era, these are directly linked to sovereignty, security, and individual rights. AI law is, before we know it, also infrastructure law.

And infrastructure law often determines people's freedom in invisible ways. Which data is shared, which entities can access it, and which logs are kept. The substance of rights protection resides more in architecture than in legal provisions. Since we are building a box called a sovereign cloud, we may not be able to avoid the constitutional law (allocation of authority and control) of that box.

6 The AI-fication of Public Services is a 'Redesign of Administrative Procedures'

The PBIA emphasizes using AI to improve public services, and in some cases, even considering personalization. This opens up a minefield that is interesting from the perspective of administrative law. When the government uses AI, the problem is not just 'is the AI correct?' The core of the procedure is called into question: 'who is considered to have made the decision?', 'how will reasons be provided?', 'can it be used as a premise for adverse actions?', and 'where will correction and relief be located when errors occur?'

Moreover, the public sector is also a massive purchaser. How procurement standards are designed changes private sector development incentives. For example, if a design that allows for audit logs is included in procurement requirements, transparency becomes a market standard rather than a 'request.' In other words, administrative law and procurement are the shortest paths to AI governance.

7 How to Read the 0.4% Figure

What is interesting is that in the PBIA's investment allocation, 'regulation and governance support' is set at 0.4%. Intuitively, this seems low. However, this is not a simple matter of good or bad. Governance becomes weak if confined to a 'fifth axis,' but becomes strong if inserted as a cross-cutting element across all axes. If the PBIA is aiming for the latter, the 0.4% is merely the 'budget for a dedicated department,' and one can read it as actual governance being embedded as an implementation requirement for other axes.

Admittedly, while cross-cutting is a beautiful philosophy, in reality, it can also become an excuse for having no one in charge. Ultimately, who performs risk assessment, who designs the audits, and who institutionalizes learning after accidents? To make cross-cutting work, you need personnel and authority dedicated to that cross-cutting role. This will likely be the part most tested during the implementation phase of the plan.

8 'Information Integrity' and the Foothold of Democracy

The reason Pillar V explicitly highlights 'democracy' and 'information integrity' is that the premises of governance are being shaken in the era of generative AI. Election interference, disinformation, phishing, synthetic voice fraud—these appear to be technical issues, but they are actually about the maintenance of the political community. What is important here is not to reduce the issue of freedom of expression to a binary choice of 'censorship or laissez-faire.' Rather, the key lies in establishing social infrastructure that guarantees authenticity, namely logs, signatures, traceability, and mechanisms for explanation.

In that sense, the PBIA's proposal for a national center for algorithmic transparency and trustworthy AI, aiming to reduce risks and ensure public trust through audits and evaluations, is not merely research and development, but also a maintenance task for democracy. Transparency is not a slogan. It is an operational design that includes personnel, procedures, measurement metrics, audit targets, and even the drawing of lines for 'areas that cannot be disclosed.' As the design of the center progresses, the issues that legal scholars should enter into will also increase. The legal nature of standards, administrative discretion, duty to explain, trade secrets, whistleblowing, and independence. Technology and law are finally being brought to the same table.

Random Thoughts

Reading the PBIA, it becomes clear that the relationship between AI and law is a matter of priority—'what does the state want to protect and what does it want to create?'—before the design of individual rules. AI is fast, and law is slow. However, the plan builds a foothold between speed and slowness for society to walk on. It is too late to build the foothold after the fact. The strength of the PBIA lies in the fact that it is trying to build the foothold first, beyond the debate of good and evil. When discussing AI governance in Japan, it may be necessary to ask 'which foothold are we building?' before entering into discussions about legal provisions.

(Magazine) 'AI and Law - Random Thoughts'

*Please refer to the following for the table of contents

note General Terms and Conditions Article 3, Paragraph 2, First Sentence
3.2 The copyright of digital content created by the creator belongs to the creator.