A status quo of global AI governance, inspired by my “Artificial General Intelligence and Human Intelligence: the Real Gap” * (Link & Video at the bottom)
Opening
Imagine a future AI assistant that does not only answer your questions. It books your doctor’s appointment, negotiates your insurance premium, selects the school materials for your child, drafts your work messages, monitors your sleep, and gently corrects your mood when you are becoming ‘unproductive.’
At first, it feels like help. Then it becomes convenience. Then it becomes default. And finally, refusal starts to look suspicious, inefficient, or even irresponsible.
Nobody forced you. Yet you were steered.
That is the quiet place where the debate about artificial intelligence is now moving. We used to ask whether machines would become smarter than us. We still ask that question, of course. But it is no longer the only question. The more urgent question is this: who remains in command when artificial intelligence starts to act, recommend, persuade, rank, monitor, buy, sell, decide, and negotiate on our behalf?
This is where regulation enters the story. Not as an abstract legal exercise. Not as bureaucracy for its own sake. But as society’s attempt to answer a simple human question: how do we keep the steering system human-governed when machine capability becomes normal?
Why this matters
In my earlier essay on Artificial General Intelligence and human intelligence, I argued that the real gap between AGI and human intelligence is not mainly raw cognitive power. It is the human steering system: our needs, emotions, hesitation, inhibition, loyalty, shame, responsibility, care, and the ability to ask not only what works, but what we can live with.
Machines can become very competent. They can process more information than we can, detect patterns faster than we can, and soon act through digital tools with a speed no human institution can match. But competence is not the same as judgment. Fluency is not the same as conscience. Optimization is not the same as responsibility.
The new global wave of AI regulation is best understood through that lens. Governments are not simply regulating software. They are trying to decide how much power may be delegated to systems that do not carry human stakes.
The new status quo: regulation has moved from principles to implementation
By April 2026, AI governance has entered a new phase. The conversation has moved beyond ethics declarations and conference language. The European Union is implementing the AI Act. The United States is trying to shape a federal framework without adopting a European-style AI code. China is regulating through security reviews, algorithm filings, content controls, and local standards. South Korea has put a broad AI framework into force. Brazil is moving toward a rights-based AI bill. International bodies such as the OECD and the Council of Europe are trying to build common language across borders.
The result is not one global rulebook. It is a patchwork. But the patchwork has a pattern. Almost everywhere, regulators are circling the same question: how do we prevent AI from becoming an invisible steering system for human life?
Different cultures answer this differently. Europe speaks the language of rights, risk and accountability. The United States speaks more in the language of innovation, free speech, markets and federal preemption. China speaks in the language of state security, social stability, content responsibility and local control. South Korea speaks in the language of industrial trust. Brazil is moving toward a rights-based model, influenced by both constitutional protections and the European example.
These are not technical differences only. They are civilizational differences. Each framework reveals what a society fears most: loss of dignity, loss of innovation, loss of state control, loss of trust, or loss of social balance.
Europe: rights, risk and the compliance state
The EU AI Act is the most complete legal architecture in the world. It divides AI systems by risk. Some uses are prohibited. High-risk systems face strict obligations. General-purpose AI models are subject to specific duties. Transparency duties apply to certain chatbots, synthetic content and emotion-recognition systems.
The important date is 2 August 2026. That is when most AI Act rules begin to apply, including many high-risk obligations, transparency rules, innovation measures and enforcement at national and EU level. Some provisions already apply, such as AI literacy duties and prohibited practices since February 2025, while rules for general-purpose AI began applying in August 2025. Full rollout continues toward August 2027 for certain product-embedded high-risk systems.
Europe’s instinct is clear: before AI becomes too deeply embedded, make it auditable. Make risks visible. Make providers document what they build. Make deployers understand what they use. Create logs, oversight, transparency, human supervision, redress and enforcement.
This is Europe’s strength. It treats AI as a matter of public order, not only private innovation. It sees that a hiring algorithm, a medical triage system, a credit model, a biometric tool, or an educational ranking system can shape life chances. Such systems cannot be left entirely to market convenience.
But Europe’s model also has a weakness. It still assumes that human oversight can be made meaningful through processes, documentation and controls. That may work for many systems. It becomes harder with autonomous agents that act quickly, combine tools, talk to other systems, and execute multi-step tasks before a human fully understands what happened.
So Europe is strong on risk governance, but still incomplete on agentic authority. It regulates many outputs and uses. It has not yet fully answered the deeper question: what may an AI agent be allowed to do on behalf of a person, a company, a hospital, a school, or a state?
The United States: innovation first, law later — or law through many doors
The United States does not yet have a single horizontal federal AI Act comparable to Europe’s. In March 2026, the White House released a National Policy Framework for Artificial Intelligence as legislative recommendations. It is not binding law. It is a map of where the federal government wants Congress to go.
The American approach is more cautious about centralized regulation. It prefers sector-specific rules, existing agencies, technical standards, sandboxes, national security tools, child protection, free speech protection, fraud prevention and targeted federal preemption of state laws considered too burdensome.
In plain English: the United States does not want fifty different state AI regimes to slow down national AI competitiveness. It wants innovation, infrastructure, investment and global leadership. It also wants to protect children, creators, consumers and speech, but without building a large AI bureaucracy.
This has real advantages. It is faster. It allows experimentation. It avoids freezing technology too early. But it leaves the human steering question more exposed. If there is no general federal baseline for high-risk AI, oversight depends on the sector: finance here, health there, employment somewhere else, consumer protection elsewhere again.
The result is flexible, but uneven. A person may be protected in one context and underprotected in another. An AI system may be treated as a medical risk, a consumer risk, a speech issue, a cybersecurity issue, or a competition issue depending on where it appears.
That may fit the American legal culture. But it does not yet fit the nature of autonomous agents, because agents do not respect sector boundaries. A single agent may read your emails, schedule your medication, manage your shopping, advise your investments, summarize your child’s school performance and negotiate with your insurer. The law sees sectors. The agent sees a life.
China: state steering before individual steering
China’s model is different again. It does not rely on one single AI Act. It governs AI through a layered system: cybersecurity law, data law, personal information protection, algorithm recommendation rules, deep synthesis rules, generative AI measures, security assessments, content requirements, algorithm filings, and local implementation standards.
China is also moving directly into the intimate zone of AI companions, digital humans and emotionally interactive systems. Draft rules announced in late 2025 and further draft regulation in 2026 focus on human-like AI interaction, labelling, addiction warnings, minors, consent, personal data, emotional dependence and public order.
This is important. China sees earlier than many Western systems that the next AI frontier is not only productivity. It is emotional interaction. AI will not only write reports. It will comfort, flatter, simulate intimacy, guide behaviour, and become a daily companion. That is where steering becomes subtle.
But the Chinese answer is not primarily cognitive liberty in the Western sense. It is provider responsibility under state supervision. The core concern is whether AI systems remain controllable, socially aligned and politically safe. Human steering is present, but mostly as state and provider steering, not as individual mental self-determination.
In this sense, China may be more operationally prepared for agentic and emotional AI, while being less protective of inner freedom as a personal right. It treats AI as a social and political infrastructure. That is realistic. But it also reminds us that the word ‘safety’ can mean different things in different systems: safety for the person, safety for the market, safety for the state, or safety for the social order.
South Korea: trust as industrial strategy
South Korea’s AI Basic Act entered into force on 22 January 2026. It is a major step because Korea is trying to do two things at once: become a global AI power and build trust in high-impact AI systems.
The law focuses on high-impact AI, generative AI disclosure, safety, certification, foreign provider duties and user notification. In sensitive sectors such as healthcare, finance, transport and nuclear safety, the law points toward human oversight. It also requires labelling and notification for AI-generated content.
Korea’s approach is not as heavy as Europe’s. It is more industry-oriented. It wants to create confidence without suffocating startups. That is why its enforcement path includes grace periods and guidance mechanisms.
The Korean model is interesting because it treats trust as a competitive asset. In a world full of synthetic media, deepfakes and automated decision-making, trust becomes infrastructure. Not soft decoration. Infrastructure.
Brazil: a rights-based model still becoming law
Brazil is not yet in the same position as the EU or South Korea. Its main AI bill, Bill No. 2338/2023, has moved through the Senate and remains under discussion in the Chamber of Deputies. Brazil currently relies on existing law: constitutional rights, data protection through the LGPD, consumer protection, civil liability, internet law and sector regulators.
But Brazil matters because it may become one of the most important rights-based AI jurisdictions outside Europe. Its legal culture already contains strong language around dignity, equality, data protection and social rights. If Brazil builds a mature AI framework, it could become a bridge between European-style rights regulation and Global South concerns around dependency, inequality and democratic resilience.
For the steering gap, Brazil’s promise is clear: AI must not become a tool by which poorer citizens, workers, children or consumers are classified and managed without real explanation or recourse. But the promise still needs operational teeth: institutions, enforcement, technical capacity, and clear rules for autonomous agents.
The missing layer: delegated agency
Across all these systems, one gap keeps returning. Most regulation still imagines AI as a tool used by a human. But the frontier is AI as an agent acting for a human or institution.
That shift changes everything.
A tool waits. An agent acts. A tool answers. An agent initiates. A tool supports a decision. An agent may prepare the decision, execute the decision, and explain the decision afterward in language that sounds reasonable.
This means regulation must move from output governance to authority governance. The key question is not only: was the AI output harmful? The deeper question is: who gave the AI authority, what limits were imposed, what did it access, which tools did it use, who could stop it, and who is responsible when it acted outside its mandate?
This is the legal vocabulary we still lack. We need a form of delegated agency law for AI. It should define what an AI agent may do, what it may never do, when human approval is mandatory, how authority is revoked, how actions are logged, and how responsibility remains traceable to humans and institutions.
Without that, the human steering gap will not be closed. It will simply be hidden behind a friendly interface.
Cognitive liberty: the next frontier
Cognitive liberty sounds abstract, but it is very concrete. It means the right to mental self-determination. The right not to have one’s attention, emotion, belief, desire or decision-making invisibly shaped by systems designed to know us better than we know ourselves.
Current AI laws protect pieces of this. The EU bans certain manipulative and exploitative practices. The United States speaks strongly about free speech, children, privacy and digital replicas. China addresses emotional dependence, labelling, minors and addictive services, but within a state-security frame. South Korea emphasizes disclosure and high-impact oversight. Brazil’s rights tradition could support stronger protections in the future.
But no major jurisdiction yet has a complete cognitive liberty framework for AI companions, persuasive agents, emotional dependency, synthetic intimacy, neurotechnology, political micro-targeting, or behavioural steering at scale.
This matters because the next AI risk is not only that systems will make wrong decisions about us. It is that systems may shape the field in which we make our own decisions. When the menu is curated, the mood is nudged, the fear is amplified, the comfort is simulated, and the default is hidden, freedom can remain formally intact while becoming practically weakened.
That is the deepest danger: not coercion, but soft capture. Not a command, but a default. Not censorship, but invisible selection. Not force, but dependence.
A practical minimum package
If we want to keep the steering system human-governed, the minimum governance package is not mysterious. It can be stated plainly.
First, separate tools from agents. A chatbot that answers a question is not the same as an agent that can send messages, spend money, access personal files or negotiate with third parties.
Second, define authority. Every agent should have clear permission boundaries: what it may read, write, send, buy, schedule, decide or trigger.
Third, require human gates for high-impact actions. Healthcare, employment, credit, legal rights, insurance, education, policing and safety-critical decisions must not drift into automatic execution.
Fourth, protect the right to disconnect. Refusal must remain real. If opting out means losing care, work, insurance, education or social standing, consent has become theatre.
Fifth, protect cognitive autonomy. No hidden manipulation, no dark patterns, no emotional exploitation, no artificial intimacy designed to create dependency, especially for children and vulnerable people.
Sixth, require audit trails. If an agent acts, it must be possible to reconstruct what it did, why it did it, what data it used, which tools it called, and who authorized the action.
Seventh, keep liability human. Machines do not carry responsibility. People and institutions do. Designers, deployers, owners and operators must remain visible in the chain of accountability.
A closing reflection
The world is not waiting for a single perfect AI law. It is building the future through imperfect rules, market incentives, national strategies, standards, court cases, platform policies and everyday defaults.
That is why the question cannot be left only to lawyers or engineers. It belongs to citizens, parents, teachers, doctors, directors, writers, workers, believers, sceptics, and anyone who still believes that human beings are more than prediction machines.
The real issue is not whether artificial intelligence will become powerful. It will. The issue is whether human beings will remain capable of steering their lives under conditions of machine persuasion, machine delegation and machine speed.
The new AI laws are not enough. But they are signals. They show that societies have begun to understand that competence without human steering is not progress. It is acceleration without orientation.
Let machines become competent. Let institutions become accountable. Let regulation become wise enough to protect refusal, dignity and thought privacy. And let the steering system remain human-governed.
Synthesis
The 2026 status quo of AI regulation shows a world moving from principle to implementation.
Europe builds a rights-and-risk compliance architecture. The United States seeks innovation and federal coherence without a full AI Act. China governs AI through operational control, security and local standards. South Korea builds trust as part of industrial strategy. Brazil is moving toward a rights-based framework.
But none of these models fully solves the central problem: AI is moving from tool to delegated actor. The next frontier is not only regulating outputs, but governing authority, cognitive liberty and the right of humans to remain in command.
Reasoning
The essay translates a legal and technical inventory into a human-facing argument. The core bridge is the earlier thesis that the real AGI gap is not intelligence alone, but steering. I treated each jurisdiction not only as a legal system, but as a cultural answer to the same question: who controls the priorities, limits and consequences of machine action?
© Robert F. Tjón, June 2026 | Creative Commons CC BY-NC-ND 4.0 International
This piece first appeared on Substack. I republish it here voluntarily — not as repetition, but as a trace; a place where words can rest after their first flight.
👉🏻 rftjon.substack.com, with video on the bottom of the article:
AI Disclosure: I utilize artificial intelligence for data synthesis and structural mapping. Please note that AI-generated insights may occasionally contain inaccuracies; readers are encouraged to verify critical information. All conclusions and final editorial decisions remain the original work of Robert F. Tjón. Content is subject to the ©️ CC BY-NC-ND 4.0 INTERNATIONAL license.
