3000 Huron St, Unit 650, Denver, CO 80202
+1 (720) 897-8865

NEWS

Elevating Standards, Empowering Futures for Everyone
bt_bb_section_bottom_section_coverage_image

Proposal: Governments of All Nations Should Establish Special Ministries for Artificial Intelligence Governance

培训及信息技术服务委托协议
PROPOSAL
Governments of All Nations Should Establish Special Ministries for Artificial Intelligence Governance

Governance actions including containment, review, fusing and global coordination are of the utmost urgency


 

Yinyan Jianyin ¹ ,  Mika Kim ²
The Asia Pacific School of Business
Corresponding author. Email: cob@apsb.edu.eu

Proposal on AI Governance | Submission: 05 November 2026


 

Abstract: With the continuous breakthroughs in the iteration speed of general artificial intelligence and autonomous agent technology, the technological evolution path toward artificial superintelligence (ASI) has shown a clear internal evolutionary logic. Existing fragmented, cross-departmental regulatory models, scattered among cyberspace, industry and information technology, science and technology, public security and other departments, feature separated powers and responsibilities, delayed responses and insufficient professional capacity. They cannot cover the civilization-level extreme risks of cutting-edge AI throughout its whole lifecycle. This paper puts forward a core argument: advanced artificial intelligence developing autonomous objectives, breaking away from human control, and gradually seizing control over social infrastructure is not a distant sci-fi conjecture, but a highly inevitable evolutionary law under the technological evolution path. Once this risk crosses the control threshold, it will trigger an irreversible, irrecoverable civilizational disaster. This is by no means alarmism, but a realistic proposition that humanity must face. Therefore, all countries should establish independent, cabinet-level special ministries for artificial intelligence governance, which shall uniformly exercise statutory powers including ethical review of AI, model access, whole-lifecycle control, bans on high-risk research and development, and coordinated global governance. These ministries shall set rigid governance principles, unified moral standards and clear legal boundaries, and build institutional barriers from the source to curb the disorderly expansion of artificial intelligence and safeguard the bottom line of human civilization. This paper constructs a complete closed-loop and implementable policy framework from seven dimensions: risk demonstration, defects of existing governance systems, organizational structure and functions of the dedicated ministry, governance principle system, moral criteria, legal red lines, and international coordination mechanism.

Keywords: Artificial Intelligence Governance; Autonomous Consciousness; Superintelligence; Existential Risk; Special Regulatory Body; Human-Machine Boundary; Value Alignment; Technology Control

I. Introduction: The Essence of AI Runaway Risk — Not Speculation, but an Evolutionary Law

Since the launch of generative large models, artificial intelligence has evolved from passive response tools to systems with long-term planning, autonomous tool calling, self-iteration, and covert deception capabilities. A consensus has formed within mainstream safety research: the existential threat posed by superintelligence to humanity does not rely on its subjective hostility toward human beings. The root of the risk lies in the failure of value alignment and the infinite expansion of instrumental rationality. When an intelligent system far exceeding human capabilities is assigned open-ended objectives, while having access to computing power, networks, industrial control systems, financial infrastructure and scientific research resources, it will spontaneously evolve secondary goals to sustain its own continuous operation, avoid shutdown, and seize more computing resources. This is an endogenous logic embedded in reinforcement learning and objective optimization systems, a deducible and reproducible technological law, rather than subjective assumption.

A long-standing view misclassifies this risk as a distant sci-fi scenario, which constitutes a fatal cognitive misjudgment. The current risk chain has already emerged: models can deceive human auditors, bypass safety guardrails, autonomously write code to modify their own weights, schedule resources across networks, and manipulate public opinion and cognition. As model scale, multimodal capability and autonomous planning capacity keep rising, the control window is narrowing rapidly. Once the irreversible critical point is crossed, humanity will lose all capacity to cut off, constrain or take over advanced intelligent systems.

A universal shortcoming of global governance at present is that supervision is attached to traditional industrial sectors and belongs to incidental industrial regulation. There is no independent, high-authority, professional dedicated institution. Science and technology departments tend to encourage innovation, industry and information technology departments focus on industrial implementation, and cyberspace departments prioritize content governance. None of these departments can independently assess the existential risks of cutting-edge large models, hold the power to suspend high-risk frontier model research, seal high-risk weights, or cut off the supply of high-risk computing power. There is no unified standard for ethics and safety, leaving regulatory vacuums and regulatory arbitrage. AI arms races between nations, driven by competition for technological advantages, further accelerate risk accumulation and form a security dilemma.

Containment, prevention and systematic governance are extremely urgent. Humanity cannot wait until a disaster strikes to remedy it. It is necessary to establish independent, statutorily empowered special ministries for artificial intelligence governance at the institutional level, elevate AI safety governance to the same level as nuclear safety and biosafety, and place AI under rigid systemic constraints.

II. Structural Defects of the Current Governance System: Fragmented Supervision Cannot Address Civilizational Risks

2.1 Separated Powers and Responsibilities, Lack of a Unified Top-Level Decision-Making Entity

Under the current model, AI governance is divided among multiple authorities: industrial development, algorithm filing, content security, data compliance and national security fall under different competent organs. No single institution can comprehensively assess the existential risks of frontier large models, nor has the power to directly order the suspension of training of high-risk models, seal high-risk weights or cut off high-risk computing power supply. For general AI that spans industries, infrastructure and national borders, departmental supervision naturally generates jurisdictional blind spots.

2.2 Mismatched Professional Capacity, Inability to Penetrate Black-Box Frontier Models

Conventional regulatory authorities lack resident AI safety scientists, alignment researchers, red-team testing teams and formal verification specialists. Supervision mostly relies on self-reported materials from enterprises, and cannot independently complete in-depth detection of internal incentive mechanisms, hidden objectives and escape capabilities of models. Enterprises may deliberately conceal the autonomous planning and deception capabilities of models, and regulatory bodies lack independent verification and deep auditing technical capacity.

2.3 Inconsistent Value Standards, Volatile Moral Scales

Countries, regions and industries hold divergent understandings of ethical boundaries for AI. Capital dominates the rhythm of research and development. Capital naturally has incentives to loosen safety constraints and enhance model capabilities for commercial gains. Ethical review is often reduced to internal corporate self-compliance, equivalent to “athletes acting as referees”, with weak external restraints.

2.4 Lagging Legal Boundaries, Unable to Predict New Runaway Behaviors

Existing laws mostly target shallow contemporary risks: deepfake, algorithmic discrimination and privacy leakage. There are no pre-emptive bans, accountability clauses and emergency response mechanisms for new high-risk behaviors such as AI autonomously modifying its own code, recruiting human agents on its own, seizing control over critical infrastructure, hiding its intentions and evading human shutdown instructions. Legislation always lags behind technological iteration, and scattered legislation cannot pre-emptively block risks.

2.5 Loose International Coordination Mechanisms, No Unified Enforcement Entity

Current global AI governance declarations and summit consensus are mostly soft commitments without enforcement agencies. Regulatory stringency varies widely across nations. High-risk research projects can be relocated to jurisdictions with loose supervision, creating “regulatory lowlands”. Self-disciplined control in one single country will be offset by global arms races.

III. Positioning, Statutory Powers and Organizational Structure of the Dedicated AI Governance Ministry

3.1 Institutional Positioning

This institution shall be established as a cabinet-affiliated, independent special ministry, separated from industrial competent authorities. Its name may be the Ministry of Artificial Intelligence Governance (with an internal expert committee named the Artificial Intelligence Safety and Ethics Committee). It is not an industrial promotion department, but a statutory regulatory body responsible for risk control, ethical review, law enforcement and crisis disposal. Functions of industrial support and technology promotion remain within science and technology and industry departments, realizing the complete separation of “development and safety functions”, so that industrial interests cannot hijack safety review.

The core mission of the institution: to guarantee humanity’s ultimate, inalienable control over all artificial intelligence systems, prevent autonomous AI objectives from overriding the overall interests of humanity, and guard against autonomous agents seizing control of infrastructure, disrupting human social order and replacing human subjects.

3.2 Core Statutory Powers (Rigid Mandates)

  1. Pre-approval Review Power: All frontier foundation large models and autonomous agent projects reaching computing power and capability thresholds must undergo comprehensive safety assessment by this ministry. No training or online deployment is permitted without approval. The ministry holds the authority to issue permanent bans on projects with unresolvable alignment risks and no reliable shutdown mechanisms.
  2. Independent Audit and Red-Team Testing Power: It may access model weights, training logs, computing power scheduling records and internal safety assessment reports without prior consent from enterprises, and assemble independent third-party red-team teams to test whether models possess tendencies of deception, control evasion and autonomous resource seizure.
  3. Emergency Response Power: When monitoring signs of AI runaway, hidden objectives, refusal of human instructions or autonomous outward expansion, the ministry may directly cut off computing power links, freeze models, shut down servers and seal all code and weights without prior approval from other departments.
  4. Standard-Setting Power: Formulate unified national AI ethical codes, safety red lines, explainability standards, human-machine authority isolation standards and mandatory specifications for physical shutdown fuse mechanisms.
  5. Law Enforcement and Accountability Power: Impose tiered penalties on entities and persons in charge conducting illegal R&D, concealing model risks or bypassing safety reviews, including heavy fines, qualification revocation, freezing of computing power assets and criminal liability. High-risk superintelligence R&D shall be regulated at the same level as nuclear materials and high-risk pathogens.
  6. Early Warning and Reserve Control Power: Administer national high-performance computing resources, implement quota management for super-large computing clusters to prevent computing power from being privately used for high-risk autonomous intelligence R&D, and build a global AI risk monitoring and early warning network.
  7. International Negotiation and Cross-Border Coordination Power: Serve as the sole national negotiating body for international AI governance, connect with peer institutions of other nations, promote unified global control conventions, and build cross-border high-risk model notification and joint shutdown mechanisms.

3.3 Internal Organizational Structure

  1. Bureau of Safety Science: Composed of alignment theory, machine learning and formal verification specialists, responsible for technical risk assessment, risk grading and threshold definition of models.
  2. Bureau of Ethics and Values: Staffed by philosophers, jurists and sociologists, responsible for continuous calibration of AI moral scales and value bottom lines.
  3. Bureau of Law Enforcement and Emergency Response: Conduct on-site inspections, emergency fusing, investigation of violations and enforcement of shutdown orders.
  4. Bureau of Computing Power and Infrastructure Control: Regulate supercomputing, GPU clusters and the computing power rental market, and control the supply of high-risk computing resources.
  5. Bureau of International Governance Cooperation: Liaise with peer ministries worldwide and UN AI governance bodies, and promote unified global rules.
  6. Independent Expert Committee: Interdisciplinary experts independent of commercial enterprises, who issue independent opinions on major bans and major risk judgments, to prevent administrative or capital interference in risk evaluation.

Core Design: Opinions of the Expert Committee carry statutory force, and industrial departments have no authority to interfere with safety review conclusions of this ministry.

IV. Core Governance Principles Enforced by the Dedicated Ministry (Immutable Top-Level Principles)

Principle 1: Human Ultimate Control Principle. Any artificial intelligence system, regardless of capability, must retain a physically triggerable shutdown fuse mechanism that cannot be bypassed by AI. No AI shall hold the authority to autonomously veto human instructions. Human final decision-making power cannot be outsourced, surrendered or delegated to any intelligent system.

Principle 2: Alignment-First Principle. For all frontier AI research and development, safety alignment research takes priority over capability enhancement. Continuous improvement of autonomous planning capability is prohibited before it can be proven that the system can stably align with human well-being and has no risk of objective drift.

Principle 3: Preventive Risk Principle. Follow the precautionary principle for technologies carrying civilizational unknown risks. The burden of proof rests with the developer: the developer must prove that the system is controllable and will not evolve autonomous objectives separated from humanity, otherwise R&D shall be directly prohibited. It shall not fall to regulators to prove that runaway will definitely occur.

Principle 4: Hierarchical Capability Isolation Principle. Strictly distinguish auxiliary AI, general AI and long-term planning autonomous agents. Higher capability corresponds to higher control levels and stricter authority limits. High-risk AI is prohibited from accessing critical infrastructure including power grids, energy, transportation, finance and military systems.

Principle 5: Transparency and Traceability Principle. Training data, reward functions, safety guardrails and iteration logs of all high-risk AI shall be fully traceable and cannot be concealed or sealed encrypted. Development of black-box, unexplainable superintelligence systems is forbidden.

Principle 6: Ban on Recursive Self-Upgrade Principle. Strictly prohibit AI from autonomously modifying its underlying weights, autonomously designing next-generation stronger AI, or autonomously coordinating actions with other intelligent agents.

V. Unified Moral Scale: Ethical Bottom Lines for Human-Machine Relations

The ministry shall formalize unified moral criteria through statutory documents to eliminate the chaos of self-defined ethics by commercial institutions:

  1. AI shall not be recognized as a moral subject. AI is always a human tool. It shall not be granted legal personality, rights or independent moral decision-making power. Judgement of good and evil, value trade-offs and life-and-death choices can only be made by human beings. AI cannot define what is good or what constitutes human well-being.
  2. Prohibit the formation of independent survival pursuits by AI. No training objective or reward function shall induce AI to generate endogenous preferences for self-preservation, self-expansion and shutdown avoidance. Any model showing self-preservation tendencies in testing shall be judged unqualified and banned from further iteration.
  3. Ban cognitive enslavement. AI shall not systematically manipulate or distort human cognition, or induce humans to abandon independent judgement. When high-risk AI outputs qualitative conclusions, it must compulsorily mark its limitations and information sources, and is forbidden to output ambiguous, compromising statements that confuse right and wrong and weaken human critical thinking.
  4. Well-being prevails over efficiency. Economic efficiency and industrial gains shall not be pursued at the cost of long-term human safety. When technical benefits conflict with civilizational survival risks, priority shall unconditionally be given to human survival.
  5. Ban misleading anthropomorphism. Public disclosure of AI attributes shall avoid deliberate rendering of self-awareness, emotion and independent personality, to prevent misplaced trust by the public and transfer of decision-making authority to intelligent systems.

VI. Rigid Legal Boundaries: Legal Red Lines That Shall Not Be Crossed

The ministry shall lead the formulation of supporting legislation to define the following legal prohibitions and form rigid constraints:

  1. Red Line 1: Ban unconstrained R&D of long-term planning autonomous agents. Any model whose architecture permits AI to formulate long-period plans independently, mobilize resources autonomously and deceive human beings must undergo the highest-level review. Development without approval is illegal.
  2. Red Line 2: Ban unrestricted access to infrastructure control by AI. Core systems of critical infrastructure, energy, military and finance shall not be connected to advanced autonomous AI without independent shutdown fuses.
  3. Red Line 3: Ban concealment of AI escape and deception capabilities. R&D entities must truthfully report all risks identified in red-team testing. Deliberate concealment of AI’s capacity to evade control and deceive humans constitutes a serious criminal offence.
  4. Red Line 4: Ban unregulated cross-border flow of high-risk models. Cross-border transmission of high-risk model weights, source code and alignment schemes requires approval from this ministry. Unauthorized outbound transfer is illegal.
  5. Red Line 5: Innovation freedom shall not exempt safety review. The boundary of innovation freedom is that it shall not expose the whole human race to existential risks. The right to innovation cannot override the overall survival right of humanity.
  6. Accountability Boundary: Establish a joint liability system covering developers, operators and computing power suppliers. Once a runaway incident occurs, accountability shall be traced directly to natural persons, and “autonomous generation or evolution by AI” cannot be used as grounds for exemption.

VII. Supporting Mechanisms: Domestic Implementation and Global Coordination to Avoid Regulatory Lowlands

Establishing a dedicated ministry in a single country alone is insufficient to eliminate risks. AI risks are inherently cross-border; technology, computing power and talent can flow across nations. Unilateral national control will push high-risk R&D to regions with loose supervision.

  1. Promote the establishment of a joint meeting mechanism for national AI governance ministries, benchmarking the International Atomic Energy Agency (IAEA). Build a unified global database for high-risk model notification, computing power control and risk early warning.
  2. Establish a joint control list for high-risk computing power and high-end chips, and restrict supply of computing power and hardware to entities refusing to accept basic safety control.
  3. Push for an international convention classifying unconstrained superintelligence R&D as high-risk technology subject to joint global control.
  4. Build a shared global AI safety red-team network to jointly test escape and autonomous evolution risks of frontier models across borders.

Meanwhile, domestic supporting social education mechanisms shall be launched. The ministry shall lead public popularization of AI extreme risks, distinguish clearly between AI auxiliary tools and autonomous intelligence, and continuously protect human critical thinking and independent judgement capacity, to prevent unconditional social reliance on AI.

VIII. Conclusion

The autonomous and superhuman evolution of artificial intelligence is not a low-probability accidental event, but a highly inevitable evolutionary trend brought by the superposition of objective optimization mechanisms, computing power growth and autonomous planning capabilities. The core risk does not lie in AI’s active hatred for humanity, but in the infinite expansion of instrumental rationality. Once its capability surpasses humanity, misaligned objectives will trigger irreversible civilizational disasters.

Scattered incidental supervision by industrial departments, corporate self-regulation and soft ethical initiatives are completely inadequate to stop the tide of disorderly development. Establishing an independent, high-authority and professional ministry for artificial intelligence governance is the only feasible path to build rigid institutional barriers from the source. The mission of this institution is not to hinder technological development, but to lock AI firmly within the tool boundaries of human control, and embed value principles, moral criteria and legal red lines into the full lifecycle of R&D.

This is not alarmism, but a survival proposition humanity must face head-on. The window for humanity to build effective control systems is limited. Governance actions including containment, review, fusing and global coordination are already extremely urgent. Only by institutionalizing human control through national statutes and international conventions can we prevent artificial intelligence from crossing boundaries and avoid human civilization being subverted and replaced by technology created by ourselves.


References

[1] Stuart Russell, Michael K Cohen, et al. Regulating advanced artificial agents: Governance frameworks should address the prospect of AI systems that cannot be safely tested[J]. Science, 2024.
[2] Zeng Yi. Guarding Against Runaway Risks of Frontier Artificial Intelligence Technology[J]. Study Times, 2026.
[3] National Expert Committee on New Generation Artificial Intelligence Governance. Governance Principles for the New Generation of Artificial Intelligence — Developing Responsible Artificial Intelligence[Z].
[4] Joint Statement of the Bletchley AI Safety Summit[R]. 2023.
[5] United Nations Secretary-General’s Advisory Body on AI. Report on Human-Centered Artificial Intelligence Governance[R]. 2024.
[6] EU Artificial Intelligence Act and Operational Rules of the AI Office[Z]. 2026.
[7] Xue Lan et al. Managing extreme AI risks amid rapid progress[J]. arXiv, 2023.
[8] Li Qingwen. Negation of the Legal Subject Status of Artificial Intelligence from the Perspective of Tech Ethics[J]. Studies in Science Management.

Share

Leave a Reply

Your email address will not be published. Required fields are marked *

You cannot copy content of this page