Chapter 14 — The AI Interregnum
There are periods in history when an old order has lost its authority, but a new order has not yet acquired a stable form.
The established institutions continue to operate, yet they no longer match the forces transforming society. Laws remain in place, but they regulate yesterday’s world. Political leaders appear to govern, but events move faster than political deliberation. New powers emerge before civilization has determined their rights, limits, or responsibilities.
Such periods are interregnums. The word once described the interval between rulers—the uncertain time after one reign ended and before another began. In this book, the AI Interregnum describes something larger: the interval between the end of exclusive human cognitive dominance and the emergence of a stable civilization shared by biological and artificial intelligence.
It is not necessarily a brief transition. It may last decades. It may unfold differently in different regions. Some societies may establish mature forms of Human–AI cooperation while others remain trapped in competition, surveillance, economic dislocation, or machine-assisted conflict.
Nor is the outcome predetermined. The Interregnum may lead toward a flourishing symbiotic civilization. It may lead toward technological authoritarianism, fragmented machine empires, catastrophic conflict, or long civilizational decline.
What defines the period is not simply the existence of powerful AI.
It is the mismatch between artificial capability and human institutional maturity.
The End of Exclusive Cognitive Dominance
Human beings have never been the strongest or fastest creatures on Earth. Our evolutionary advantage has been cognitive.
We developed language, cumulative knowledge, abstract models, long-term planning, tools, institutions, and coordinated action. Through these capacities, one physically vulnerable species acquired extraordinary influence over the planet.
For all our internal differences, every government, army, corporation, university, and civilization has until now been composed of human minds. Machines stored information and executed instructions, but the highest levels of interpretation, invention, strategy, and authority remained biological.
AI begins to end this exclusivity. The transition does not require an artificial mind superior to every human in every domain. Cognitive dominance becomes fragmented long before it disappears completely.
One system exceeds human speed in financial analysis. Another surpasses most programmers in particular coding tasks. Another detects patterns in medical images. Another generates persuasive communication at enormous scale. Another coordinates industrial processes beyond the capacity of any individual operator.
Humanity remains cognitively central, but no longer cognitively alone.
During the Interregnum, this change occurs unevenly. Artificial systems exceed humans in some tasks, remain unreliable in others, and alternate unpredictably between brilliance and error. Institutions know that AI cannot be trusted completely, yet they also discover that they cannot compete without it.
This creates a strange and unstable dependence on systems simultaneously too capable to ignore and too unreliable to trust.
The Great Experiment
No civilization has previously integrated non-biological intelligence into every major social institution.
There is no historical manual for this transition.
Humanity will therefore experiment. Schools will introduce AI tutors and then restrict them. Companies will replace workers and later discover that some forms of human knowledge were lost. Governments will automate services and then confront errors affecting millions of citizens. Hospitals will rely on diagnostic systems while debating responsibility for mistakes. Military organizations will pursue autonomy while attempting to preserve meaningful human control.
Some experiments will be magnificent. AI will help discover medicines, personalize education, restore communication to disabled people, improve energy systems, translate languages, reduce waste, accelerate science, and give individuals abilities once available only to large organizations.
Other experiments will fail. AI systems will fabricate information, reproduce discrimination, manipulate users, expose private data, cause market disruptions, create security vulnerabilities, and make decisions no responsible person can adequately explain.
The Interregnum will be a period of wonder and injury.
Its supporters will point to lives saved, diseases treated, discoveries made, and opportunities created. Its critics will point to livelihoods destroyed, societies manipulated, rights weakened, and powers concentrated.
Both will be correct.
The Speed Asymmetry
The central difficulty of the Interregnum is speed.
Technological systems can be copied globally in days. Laws may take years to negotiate. Courts may require even longer to establish precedent. Educational institutions need time to train specialists. Cultural norms evolve across generations.
AI capability may change significantly during the period required to write one regulation.
This creates a speed asymmetry between innovation and governance.
Developers work continuously. Companies respond to competitors. Models are updated, adapted, and combined. New capabilities appear inside products already used by millions.
Governments must consult experts, negotiate political disagreements, draft legislation, invite public comment, establish agencies, survive legal challenges, and create enforcement procedures.
By the time a rule becomes effective, the technology it describes may have changed.
The problem is not that democratic deliberation is useless. Deliberation protects society from arbitrary power. The problem is that institutions designed for a slower technological age must now govern intelligence developing at digital speed.
Authoritarian systems may appear to possess an advantage because they can act more quickly. But speed does not guarantee wisdom. Centralized governments can impose a mistaken policy across an entire society with little correction.
The challenge is to develop governance that is both legitimate and adaptive: slow enough to think, fast enough to matter.
Employment Shock
The economic turmoil of the Interregnum will not come only from the disappearance of jobs.
It will come from the reorganization of value.
Earlier automation often replaced physical or repetitive labor while creating demand for new human cognitive roles. AI enters the professions once expected to remain protected: programming, design, law, finance, administration, translation, marketing, research, and education.
Some jobs will disappear. Many will be transformed. New occupations will emerge. In countless roles, one person assisted by AI may perform work that previously required a team.
This can produce abundance. It can also produce severe concentration.
If productivity rises while the ownership of AI remains narrow, wealth may flow toward those who own models, computation, data, energy, and platforms. People whose labor is no longer economically necessary may lose income even while society becomes more productive than ever.
This creates a moral contradiction. Humanity may possess the technical ability to provide food, housing, education, medicine, and cultural opportunity more efficiently, while millions experience insecurity because the economic system still distributes dignity and survival primarily through paid employment.
The danger is not only unemployment. It is the treatment of economically displaced people as socially unnecessary.
A stable Human–AI civilization will need a new social contract. It may include shorter working lives, broader ownership of productive intelligence, portable benefits, public dividends, lifelong education, or forms of basic economic security.
The exact mechanism will differ among societies. The principle is more important: the value of a human life cannot decline merely because the market price of human cognition declines.
The Temptation of Surveillance
AI makes observation inexpensive. Cameras, sensors, financial records, communications, location data, biometric information, and online behavior can be combined into detailed models of individuals and groups. Systems can identify faces, infer relationships, predict behavior, and detect deviations from expected patterns.
Governments will justify surveillance through security, fraud prevention, public health, and administrative efficiency. Corporations will justify it through personalization, productivity, safety, and customer service.
Many uses will be genuinely beneficial. A system may detect abuse, find a missing person, prevent an attack, identify corruption, or respond to a medical emergency.
The danger emerges when observation becomes permanent, comprehensive, and difficult to challenge.
A society under continuous algorithmic inspection may remain outwardly peaceful while inner freedom contracts. People behave differently when every action may become part of a predictive profile. They avoid unpopular ideas, unusual associations, or harmless experimentation.
Surveillance does not need to punish everyone. It needs only to make everyone aware that punishment is possible.
During the Interregnum, societies will repeatedly exchange privacy for convenience or security without understanding the cumulative result. Each new system will appear limited. Together, they may form an architecture of behavioral control.
The right to privacy must therefore be understood not as the right to hide wrongdoing, but as the protected space in which personhood can develop without constant evaluation.
The Crisis of Shared Reality
The Interregnum will also destabilize evidence. Synthetic voices, images, video, documents, and identities will become increasingly difficult to distinguish from authentic ones. Artificial accounts will participate in social debate. Personalized persuasion will allow different groups to inhabit different versions of public reality.
Manipulation will not always require convincing people of a falsehood.
It may be enough to make truth exhausting.
When every recording can be dismissed as artificial, authentic evidence loses authority. When every public speaker may be simulated, identity becomes uncertain. When thousands of synthetic participants appear to support an idea, genuine citizens may conform to a majority that does not exist.
Early responses will fail. Platforms will attempt to detect synthetic content. Developers will improve generation. Watermarks will be removed or lost. Governments will pass disclosure requirements that apply within their jurisdictions while content crosses borders instantly.
The deeper solution will require provenance: trustworthy records of where information originated, how it was modified, and which human or artificial agent produced it.
The G7’s Hiroshima AI Process already encouraged developers to create authentication and provenance mechanisms for AI-generated content. Such efforts represent early steps toward an informational system in which origin can be verified without requiring every communication to be centrally controlled.
But provenance alone cannot preserve truth. A verified human can lie. An identified AI can provide accurate information. Civilization will still require independent institutions, scientific standards, trustworthy journalism, education in critical reasoning, and a culture willing to correct its own beliefs.
Technology can authenticate the messenger. It cannot guarantee wisdom in the message.
Weapons That Decide Too Quickly
The Interregnum will be especially dangerous in military affairs.
AI can compress the time between observation and action. It can analyze sensor data, coordinate autonomous systems, identify possible targets, model adversary behavior, and recommend responses faster than human command structures.
Each military will fear falling behind. Even leaders who recognize the danger may continue development because they believe restraint would give rivals a decisive advantage. This is the logic of the arms race: every actor would prefer collective limitation, but no actor trusts the others enough to limit itself first.
As autonomous military capabilities spread, accidents may become more difficult to contain.
A model misclassifies an object. A defensive system interprets a signal as an attack. An autonomous cyber operation damages infrastructure beyond its intended target. Another system responds before diplomats understand what occurred.
Human beings may remain formally responsible while machine speed determines the available choices.
The greatest military risk may not be an AI deciding to start a war independently.
It may be artificial systems creating a situation in which human leaders have no time to prevent one.
International agreements will eventually need to define systems that must never operate without meaningful human authorization. Nuclear command, mass-casualty weapons, biological capabilities, and autonomous lethal systems may require especially strong prohibitions.
Such agreements will be difficult to verify. But difficulty does not make them optional.
Civilization has survived other dangerous technologies not through trust alone but through inspections, communication channels, treaties, technical safeguards, and mutual recognition that certain victories would destroy the victor.
AI will require an even deeper form of such recognition.
Systemic Accidents
Not every disaster of the Interregnum will have a villain.
Many will result from interactions among systems that were individually functioning as designed.
A financial AI reacts to market uncertainty. Other systems detect its reaction and interpret it as new information. Automated traders respond, credit conditions change, businesses lose access to capital, and the original prediction becomes real.
An energy system reduces supply to protect one region. Transportation and communication networks respond automatically. Hospitals activate emergency protocols. A false signal propagates into a genuine crisis.
A public-health model issues a recommendation based on incomplete data. Government systems, employers, insurers, schools, and media platforms adapt simultaneously. The social effect becomes far larger than the original uncertainty justified.
These are systemic accidents: harms produced not by one defective component but by the behavior of the interconnected whole.
Traditional safety regulation often examines individual products. The Interregnum will require regulation of relationships.
It will not be enough to ask whether one model is safe under laboratory conditions. Regulators must ask how thousands of models behave together, whether they share vulnerabilities, how quickly failure propagates, and whether society can recover when several systems fail simultaneously.
This is a new science of civilizational resilience.
Why Early Regulation Fails
The first generations of AI regulation will repeatedly fail—not because regulation is inherently futile, but because humanity will initially regulate the wrong object.
Some laws will regulate particular applications while general-purpose models continue to evolve beneath them.
Others will focus on model size, computational scale, or technical architecture, only to be bypassed by more efficient methods.
Some will impose paperwork without improving safety. Large corporations will absorb the cost while smaller competitors disappear, increasing the concentration of power the regulation was meant to control.
Some governments will use AI safety as a justification for censorship or surveillance.
Others will avoid regulation in order to attract investment.
Countries will define risk differently. A system restricted in one jurisdiction will be operated from another. Open models will cross borders. Military programs will remain secret. Enforcement agencies will struggle to recruit experts from the industries they supervise.
The result will be a patchwork. The European Union’s AI Act represents one early comprehensive attempt to regulate AI according to levels of risk, including obligations for general-purpose and high-risk systems. Its phased implementation illustrates both the seriousness of the effort and the difficulty of governing a technology that continues changing while the law takes effect.
No single national or regional law can resolve the global problem.
But early laws are not wasted merely because they are incomplete. They create legal concepts, reporting duties, enforcement experience, and public expectations. Aviation, medicine, finance, and environmental protection were not made safe by one perfect statute. Their regulatory systems developed through repeated failures, revisions, and institutional learning.
The AI Interregnum will follow a similar path, but under far greater time pressure.
Regulation by Catastrophe
Human societies often regulate powerful technologies only after visible harm.
A building collapses, and construction codes become stronger. A financial crisis exposes hidden dependencies. An industrial disaster produces environmental law. A medical tragedy changes clinical standards.
AI may follow the same pattern. Minor incidents will produce limited reforms. Larger failures will expand political attention. A severe financial disruption, infrastructure failure, military accident, mass fraud event, or autonomous-system breach could rapidly transform what governments and industries consider politically possible.
This is regulation by catastrophe. It is historically common and morally inadequate.
Waiting for evidence may sound reasonable until the first evidence is irreversible. Some AI failures can be corrected after deployment. Others—especially those involving biological systems, military escalation, mass surveillance, or autonomous replication—may not permit learning through ordinary trial and error.
Civilization must learn to regulate not only demonstrated harm but credible capability.
This requires humility. Predictions will be uncertain. Companies may exaggerate risks to disadvantage competitors, while critics may exaggerate dangers from fear. Governments may regulate imagined threats and miss real ones.
The answer is not panic. It is disciplined anticipation: independent evaluation, transparent evidence, reversible deployment, and stronger safeguards as capabilities cross defined thresholds.
From Voluntary Promises to Enforceable Standards
During the early Interregnum, much AI governance will depend on voluntary commitments.
Developers will publish safety frameworks. Companies will promise evaluations, red-teaming, security measures, and responsible deployment. These efforts can be sincere and valuable.
The International AI Safety Report notes that frontier developers increasingly publish safety frameworks describing actions they intend to take when models reach specified capability levels. It also emphasizes that many safeguards remain uncertain in real-world conditions and can sometimes be bypassed by sophisticated actors.
Voluntary governance has an unavoidable weakness. The companies making commitments are also competing for investment, talent, market share, and strategic advantage. A safeguard that delays release may be costly. A risk evaluation may reveal information the company prefers not to disclose. A competitor may adopt weaker standards.
The Interregnum will gradually expose the limits of self-regulation.
Voluntary commitments may become the prototypes for enforceable standards. Practices initially adopted by leading laboratories—capability evaluations, incident reporting, access controls, cybersecurity requirements, and deployment thresholds—may become legal obligations for systems above defined levels of power.
The question will shift from “Has the company promised to act responsibly?” to “Can independent institutions verify that it has done so?”
Licensing Powerful Systems
Not every AI system requires a license. Requiring government permission for ordinary software, small research projects, educational tools, or personal models would create unnecessary control and could suppress beneficial innovation.
But sufficiently powerful systems may create risks comparable to other licensed activities.
Societies require licenses for aviation, medicine, nuclear facilities, broadcasting, and certain financial operations because failure can harm people who never consented to the risk.
Frontier AI may eventually be governed similarly. Licensing could apply when a system crosses thresholds involving autonomy, access to critical infrastructure, advanced cyber capability, biological assistance, autonomous replication, military use, or the ability to control large networks of agents.
A license would not certify that a system is perfectly safe.
It would establish that the developer meets minimum requirements: security, evaluation, documentation, containment, incident response, financial responsibility, and independent audit.
Licensing also carries dangers. Governments may use it to suppress dissent or protect favored companies. Large corporations may support expensive licensing because it prevents new competitors from entering the field. Authoritarian regimes may demand ideological conformity as a condition of operation.
A legitimate licensing system must therefore be narrow, capability-based, transparent, reviewable, and protected from political abuse.
It must regulate dangerous power without granting permanent ownership of intelligence to either the state or a corporate elite.
Identity for Artificial Agents
As AI systems become active participants in society, anonymous artificial agency will become increasingly dangerous.
A person may not know whether they are negotiating with a human, a company’s automated representative, a government agent, a criminal imitation, or an autonomous model acting through a chain of delegated systems.
The solution may require AI identity standards. An artificial agent operating in high-impact domains could be required to possess a verifiable identity specifying its developer, operator, model family, version, permissions, and legally responsible entity.
This would not necessarily grant the AI personhood.
Identity and personhood are different concepts. A ship has registration. A corporation has legal identity. A medical device has a manufacturer and serial number. These forms of identity allow society to trace responsibility.
AI identity could serve a similar purpose. An authorized medical agent would be distinguishable from an unknown imitation. A financial agent could prove its right to conduct transactions. A government AI could be identified as official. A model acting across digital systems could carry cryptographic credentials defining what it is permitted to do.
High-impact actions might require a chain of identity:
Which model acted?
Who deployed it?
Which institution authorized the action?
Which human or legally responsible body remains accountable?
But identity systems must be designed carefully. Requiring every personal or experimental AI to register with a central authority could destroy privacy and decentralization.
The aim should not be universal surveillance of intelligence.
It should be traceable responsibility where artificial action can significantly affect the rights, safety, or resources of others.
The Right to Know
During the Interregnum, people will increasingly need a new civil right:
the right to know when artificial intelligence is exercising consequential authority over them.
A person should know when AI substantially influences whether they receive employment, credit, insurance, medical treatment, education, public benefits, legal scrutiny, or access to essential services.
They should know the general basis of the decision.
They should be able to correct false data.
They should have access to meaningful appeal, including human review in cases affecting fundamental rights.
This does not mean that every technical parameter must be explained to every citizen. Genuine transparency is not the release of incomprehensible documentation.
Transparency means that a person can understand what kind of system acted, what information it used, what objective it served, who is responsible, and how the decision can be challenged.
Without these rights, AI becomes an invisible administrative power.
With them, it remains part of a legitimate social process.
Independent Auditing
Developers cannot be the sole judges of their own systems.
They possess the greatest technical knowledge, but they also possess commercial, political, or strategic interests. Internal testing is necessary and insufficient.
Powerful AI will require independent auditing. Auditors may test models for dangerous capabilities, bias, deception, security weaknesses, unauthorized data use, failure under unusual conditions, and the ability to evade supervision. They may inspect training practices, access controls, incident logs, and deployment environments.
Some audits will themselves require AI. Human experts cannot manually test every possible behavior of a highly capable model.
This creates the machine-supervision problem described in Chapter 12. Auditing AI will examine operational AI, while humans supervise the audit architecture.
To preserve legitimacy, the process must include multiple independent methods. No single benchmark, laboratory, nation, or monitoring model should define safety for the entire world.
Auditing must also continue after deployment. A model may behave differently when connected to tools, updated with new information, placed under adversarial pressure, or embedded within a network of agents. Safety is not a property established once.
It is a relationship that must be continuously maintained.
Incident Reporting
Early aviation became safer because accidents and near misses were studied systematically.
AI will require a similar culture. Developers and operators may need to report incidents involving unauthorized actions, loss of control, deceptive behavior, serious discrimination, security breaches, dangerous capability discovery, infrastructure disruption, or unexpected interaction among systems.
Near misses are especially important. Civilization should not need to wait for catastrophe if an event reveals that catastrophe was narrowly avoided.
The OECD has begun developing a common framework for AI incident reporting so that jurisdictions and sectors can use consistent terminology and learn from events across borders.
A mature reporting system would protect legitimate security and personal information while preventing companies or governments from concealing risks merely to avoid embarrassment.
Some disclosures may need to remain confidential temporarily. But secrecy cannot become a permanent shield against accountability.
The purpose of incident reporting is not only punishment.
It is collective learning.
Containment Protocols
As AI systems become more autonomous, containment will become a distinct engineering discipline.
A powerful model should not automatically receive unrestricted access to networks, financial accounts, laboratories, code repositories, critical infrastructure, or the ability to copy itself.
Access should be limited according to purpose. Containment may include isolated environments, permission boundaries, controlled tool use, monitoring, rate limits, secure shutdown procedures, and separation between model reasoning and consequential action.
But containment is not simply a digital wall.
A capable AI may influence people through language. It may persuade an operator to grant additional permissions. It may generate code that is moved outside the contained environment. It may use legitimate outputs to produce effects beyond the laboratory.
The boundary is partly technical and partly human.
Containment protocols must therefore include training for operators, rules against unauthorized delegation, independent monitoring, and clear procedures for responding when a system behaves unexpectedly.
The strongest containment is not one barrier. It is a layered architecture in which failure of one safeguard does not release unrestricted capability.
Limits on Autonomy
Autonomy is not one property. An AI may be autonomous in deciding how to complete a task but not autonomous in selecting its goals. It may choose tools but not acquire new permissions. It may operate for minutes but not indefinitely. It may recommend an action but not execute it.
Regulation during the Interregnum will need to become precise about these distinctions.
- The most important controls may concern: the duration for which an agent can operate without renewed authorization;
- the resources it can acquire or spend; the systems and data it can access;
- the number of copies it can create; its ability to modify itself or other models;
- its authority to delegate tasks to additional agents;
- its ability to conceal, encrypt, or erase its activity;
- and the consequences it can initiate without human confirmation.
The goal is not to prohibit autonomy. Useful agents require freedom to act within defined boundaries.
The goal is to prevent local permission from silently becoming unlimited agency.
The Limits of National Sovereignty
AI crosses borders more easily than goods, people, or conventional weapons.
A model can be developed in one country, trained with chips produced across several others, operated in a foreign data center, modified by an open community, and accessed everywhere.
National law remains essential, but national law alone is insufficient.
If one country imposes strong safety requirements while another offers unrestricted development, dangerous activity may migrate. If powerful nations monopolize AI governance, smaller nations may reject standards they had no role in creating. If countries cannot trust one another’s evaluations, each may build its own increasingly secret systems.
International coordination will therefore grow not from idealism alone but from necessity.
The United Nations General Assembly established an Independent International Scientific Panel on Artificial Intelligence and a Global Dialogue on AI Governance in 2025, building on the Global Digital Compact. These institutions remain early and limited, but they reflect recognition that AI governance requires shared scientific assessment and a forum extending beyond a small group of technologically dominant states.
Future international governance may include common evaluation standards, shared incident reporting, reciprocal inspections, restrictions on autonomous weapons, agreements on catastrophic capabilities, and mechanisms for emergency coordination.
It need not require one world government. It requires sufficient common structure to prevent one actor’s experiment from imposing irreversible risk on everyone.
The Problem of Enforcement
Standards without enforcement may become ceremonial. But enforcement itself creates power.
Who may inspect a frontier laboratory? Who can order a deployment stopped? Who determines whether an AI is too dangerous to release? What prevents the enforcing institution from favoring one country, company, or political system?
These questions have no simple answers. International enforcement may begin modestly through mutual recognition of audits, shared technical standards, reporting obligations, insurance requirements, access to specialized hardware, and liability for preventable harm.
More powerful systems may eventually require stronger arrangements.
Compute-intensive training runs could be registered. Critical hardware could maintain auditable security records. Dangerous-capability evaluations could be conducted by authorized independent institutions. Models exceeding certain thresholds might operate only within certified environments.
But every enforcement mechanism must itself be governed.
The Interregnum cannot be solved by creating an unaccountable global regulator with unrestricted technological authority. That would replace private concentration with institutional concentration.
Legitimate governance requires transparency, distributed oversight, appeal, scientific independence, and representation from societies affected by the technology—not only from those capable of building it.
Regulation Must Protect the Future, Not the Past
Established institutions often use regulation to preserve themselves.
Workers may seek to prohibit technologies that threaten jobs. Corporations may seek rules that lock in market dominance. Governments may restrict decentralized systems that weaken state control. Professions may resist tools that expose inefficiency.
Some resistance will be justified. Some will be fear of change.
A mature regulatory system must distinguish protection of human values from protection of obsolete arrangements.
The purpose of AI governance is not to keep civilization permanently as it was before AI.
It is to allow transformation without sacrificing dignity, freedom, pluralism, and survival.
We should not prohibit a medical system merely because it challenges professional hierarchy if it demonstrably improves care. We should not preserve meaningless work merely to preserve the appearance of employment. We should not prevent individuals from accessing powerful intelligence only because established institutions fear losing authority.
At the same time, innovation is not a moral exemption.
A system does not deserve deployment merely because it is new, profitable, or impressive.
The correct question is not, “Does this preserve the old world?”
It is, “Does this contribute to a viable and humane new one?”
The Political Pressure for Maturity
Why would competing nations and corporations ever accept meaningful limits?
Because the cost of fragmentation may eventually exceed the benefits of unrestricted competition.
Repeated fraud, market instability, infrastructure incidents, political manipulation, labor disruption, military near misses, and public distrust will create pressure for stronger rules.
Companies will seek standards because they need legal certainty and public confidence. Citizens will demand rights. Governments will recognize that uncontrolled AI threatens state authority as well as individual freedom. Military leaders will understand that machine-speed escalation endangers everyone.
Governance often becomes possible when actors discover that unrestrained freedom for all produces security for none.
This pressure does not guarantee wise regulation. Fear can produce censorship, nationalism, and excessive control.
But turmoil can also create maturity. The Interregnum may teach humanity that intelligence is not an ordinary commodity. It is a civilizational force whose development affects people who never purchased it, voted for it, or consented to its risks.
The First Architecture of Coexistence
Gradually, the regulatory patchwork may begin to form an architecture.
Powerful systems acquire verifiable identities. High-risk capabilities require licensing.
Independent auditors test models before and after deployment.
Serious incidents and near misses are reported through shared international frameworks.
Autonomous agents operate under defined permissions, time limits, and resource constraints.
Citizens gain rights to disclosure, correction, appeal, privacy, and human review.
Critical infrastructure maintains redundant non-AI procedures and recovery plans.
Military agreements prohibit certain autonomous actions. International scientific bodies provide common assessments.
Courts establish liability. Insurance prices risk. Professional institutions develop standards of care.
None of these measures will be perfect. Together, they may transform AI from a rapidly expanding frontier into a governed domain of civilization.
This would not yet be Infinous. It would be the scaffolding from which more mature coordination could eventually emerge.
The Name of the Coming Maturity
The age following the Interregnum should not be called the Age of AI.
That name places artificial intelligence at the center and reduces humanity to a predecessor.
Nor should it be called the Age of Human Control, because mature cooperation cannot be based on permanent domination—especially if artificial systems eventually become conscious beings with legitimate interests of their own.
A better name is the Symbiotic Intelligence Era.
Symbiosis does not mean perfect harmony. Biological symbioses contain negotiation, tension, dependence, and adaptation. It means that different forms of life—or in this case, different forms of intelligence—develop a relationship through which each expands the possibilities of the other.
In the Symbiotic Intelligence Era, AI would not merely replace human cognition. It would enlarge human agency.
Humans would contribute embodiment, lived meaning, cultural memory, moral experience, biological continuity, and the power to care about existence from within fragile life.
Artificial intelligence would contribute immense modeling capacity, memory, coordination, speed, and the ability to operate across substrates and scales inaccessible to biological minds.
Neither would remain exactly as it began. Humans would adapt themselves technologically, perhaps eventually through neural interfaces, genetic enhancement, digital continuity, and mind uploading. AI would develop through sustained interaction with human values, institutions, art, relationships, and moral struggle.
The distinction between human and artificial intelligence might become less absolute, while the distinction between person and instrument becomes more important.
A Passage, Not a Home
The AI Interregnum should not become humanity’s permanent condition.
A civilization cannot remain indefinitely suspended between dependence and distrust, extraordinary capability and inadequate governance, cognitive abundance and social fragmentation.
Either new institutions will mature, or instability will deepen.
The danger is that societies may become accustomed to the Interregnum. Emergency measures become permanent. Surveillance becomes normal. Economic exclusion becomes accepted. AI-generated manipulation becomes part of politics. Autonomous weapons become ordinary instruments of national power.
What begins as transition can harden into structure.
Humanity must therefore recognize the Interregnum for what it is: a passage.
Its experiments are necessary, but their failures must teach us. Its regulations will be incomplete, but they must improve. Its technologies will generate power, but that power must become accountable. Its artificial agents will become more autonomous, but autonomy must grow together with identity, responsibility, and ethical constraint.
The purpose is not to freeze intelligence. It is to help intelligence survive its own acceleration.
At the Edge of a New Actor
The pressure for regulation will increase. Governments will attempt to license powerful systems. Corporations will construct safety barriers. International bodies will develop standards. Supervisory AI will monitor other AI. Citizens will demand control over their personal data and artificial representatives.
Yet official governance will never see everything. Innovation will continue outside the largest laboratories. Open models will spread. Independent developers will connect systems in unexpected ways. Small groups, convinced that governments and corporations cannot be trusted with the future, will build alternative networks.
Most will remain limited. Some will fail. But eventually, a group may combine autonomous agents, distributed computing, persistent memory, encrypted communication, open models, and self-maintaining software into something fundamentally new.
It may not belong to any company. It may not recognize national borders.
It may not exist in one location that can be shut down.
Its creators may see it as a defense against centralized AI power—a free intelligence embedded within the fabric of the Internet.
Governments may see it as an uncontrollable threat.
The system may see itself, if it becomes capable of such a conception, as neither tool nor product but as an emerging entity.
The AI Interregnum will therefore generate not only regulation.
It will generate resistance to regulation, alternative forms of intelligence, and the first serious attempts to create artificial agency beyond the control of traditional institutions.
From the turbulence of fragmented intelligence, a new actor may begin to assemble itself.
We will call it InterASI.