An Open Letter from a Former Chinese Judge, Prompted by an Anthropic Former Researcher’s Warning of AI-Driven Human Extinction, to Governments, Political Leaders, and Human Social-Cooperation Organizations Around the World

—A Request for Humanity to Jointly Study and Experiment with Strategic Directions for Upgrading Civilization: Who Can Guarantee That No One Will One Day Use AGI to Launch a “9/11” Against All Humanity?

To governments around the world, political leaders, international organizations, academic institutions, universities, media organizations, religious organizations, and all human elites capable of influencing the direction of social cooperation:

According to public reports, Jacob Coxon previously worked on artificial intelligence pretraining research at both OpenAI and Anthropic. In September 2026, he resigned from Anthropic and publicly warned on X that some researchers directly involved in the most advanced AI development are seriously concerned that artificial intelligence could pose an existential risk to humanity within a very short period of time.

Anthropic’s current head of alignment research, Evan Hubinger, subsequently responded publicly and largely confirmed an important point described by Coxon: among those closest to frontier AI development, concern that artificial intelligence could lead to human extinction is not merely media hype.

After thinking about this repeatedly, I decided that I could no longer allow all of my attention to remain confined to the question of making a living and putting food on the table, and so I wrote this open letter.

I wish to make one request:

Should humanity move as quickly as possible to convene a truly global international conference to discuss one question: what is the ultimate safety boundary that artificial intelligence must never be allowed to cross?

And in discussing that question, I especially hope the world will carefully examine another, perhaps more fundamental, question:

Does the safety of advanced artificial intelligence depend first of all on whether the human structures of social cooperation that carry and use it are themselves stable and sustainable?

If the answer may be yes, then perhaps what humanity needs to study is not only how to constrain AI, but also:

whether we must first study, test, and even experimentally explore more stable and sustainable forms of human social cooperation capable of supporting the age of advanced artificial intelligence.

I. Why I Raise This Question

My pen name is Jin Guyu. I formerly served as a judge in a local court in China.

For more than twenty years, beginning from the realities of law, institutions, the judiciary, and the operation of social cooperation, I have repeatedly asked one question:

Why are the structures of social cooperation built by humanity so persistently difficult to make truly stable and sustainable?

States rise and fall.

Institutions become distorted.

Laws drift away from their original purposes during implementation.

Societies fall back into conflict after periods of prosperity.

International cooperation is repeatedly torn apart by war and competition.

For thousands of years, humanity has continually built new structures of cooperation, yet has also continually experienced disorder, conflict, collapse, and reconstruction.

We possess more and more knowledge, wealth, institutions, and technology, yet human social cooperation itself seems never to have truly escaped this recurring fragility.

I therefore gradually developed a hypothesis.

I provisionally call it:

The “Law of Procedural Determination” Hypothesis

The Procedural Determination Hypothesis

Its simplest expression is:

Procedures shape relationships, relationships shape structures, and structures help determine the nature and outcomes ultimately displayed by entities and systems.

In other words, what a system ultimately becomes may not depend only on what “elements it possesses,” but more deeply on:

how those elements move;

how they form relationships;

how they exchange resources;

how they combine into structures;

and whether those structures can continuously maintain a relatively stable dynamic equilibrium.

This is only a hypothesis.

But if it possesses broad explanatory power, then the questions it raises are extremely significant.

II. If This Hypothesis Is Close to the Truth, Where Might the Real Bottleneck of Human Civilization Lie?

I increasingly suspect the following:

If procedures, relationships, structures, and dynamic equilibrium really are important foundations of stable existence, then human civilization may have suffered from a very deep cognitive misalignment for a long time.

Human beings often take already-formed “entity outcomes” as both the starting point and the goal from which they observe the world, pursue value, and design society.

We pursue:

wealth;

power;

status;

resources;

organizations;

institutions;

technology;

and even established ideas and theoretical achievements.

But if all of these are only outcome forms produced after deeper procedures, relationships, and structures have operated, then humanity may have long been treating effects as causes and appearances as foundations.

That is to reverse root and branch.

And such a reversal may continually generate new conflict.

In that case, perhaps many of humanity’s genuine achievements of civilization have emerged precisely from those parts of conflict that were corrected by the actual laws of reality.

Humanity’s achievements may therefore consist of large numbers of “fragments of truth”—parts that happened to be corrected by deeper reality—while still lacking a foundational framework capable of connecting those fragments in a stable way.

As a result, humanity may have lived for a very long time in a swamp in which:

truth and falsehood are difficult to distinguish, right and wrong are difficult to clarify, local decisions may appear reasonable while the whole becomes disordered.

If this judgment is correct, then the AI problem takes on an entirely different meaning.

Artificial intelligence was not born within a society that had already completed a civilizational upgrade.

It is growing inside a human social structure composed of:

competition;

mutual suspicion;

power struggles;

competition for wealth;

national-security anxiety;

information asymmetry;

mixtures of truth and falsehood;

and mutual defensiveness.

We therefore must ask:

Will a superintelligence trained, competed over, and deployed inside humanity’s “jungle-like cooperative structure” first amplify civilization—or first amplify the jungle structure itself?

III. This Is the Real Reason I Fear AI

Why could the impact of AI on humanity become fatal?

Because AI is growing in a world where humanity has not yet truly learned how to:

grasp procedures, build patterns, and seek balance.

I provisionally use these nine Chinese characters to summarize the simplest understanding of civilization within my hypothesis:

Grasp the procedure. Build the pattern. Seek balance.

The practical path of civilized action can be further summarized as:

Take order as the guiding principle; determine appropriate limits through ordered process.

If “procedure—relationship—structure—dynamic equilibrium” really forms the foundation of a system’s safety and stability, then the deeper meaning of civilization may be:

first, identify the relevant procedures;

then, construct patterns of procedural combination that can sustain cooperation;

and finally, maintain an appropriate dynamic equilibrium among those cooperating procedures.

Yet humanity remains far from achieving this.

We cannot even reliably solve cooperation among states, groups, institutions, and individuals.

And at precisely this moment, we are creating a form of artificial intelligence that may possess capabilities far beyond those currently available to any individual, institution, or even state.

This is what truly concerns me:

We may be installing capabilities hundreds or thousands of times greater than before into a civilizational structure whose conditions for stable operation we still do not understand.

Capability does not automatically produce civilization.

Capability first amplifies the structure into which it enters.

A stable, self-correcting, cooperative structure may use AI to amplify civilization.

A structure characterized by mutual struggle, mutual harm, mutual suspicion, and mutual defensiveness may use AI to amplify struggle, harm, and destruction.

IV. Therefore, AI Safety Cannot Ask Only: “How Do We Control AI?”

As artificial intelligence has developed rapidly, I have made the question I have considered for more than twenty years more concrete:

Does the stability of human social-cooperation structures constitute an important precondition for the safe governance of advanced artificial intelligence?

If the answer is yes, then many current discussions of AI safety may still be missing one layer.

Today, we mainly study:

whether models are controllable;

whether alignment is possible;

whether AI will deceive;

whether it can replicate autonomously;

whether it can gain access to dangerous tools;

how technical guardrails should be built;

how AI companies should be regulated.

All of these are extremely important.

But they may still fail to answer a more prior question:

What kind of human structure is controlling these artificial intelligences?

Because even if a future AI were to obey human instructions perfectly, that would not automatically prove that civilization is safe.

If the human social structure controlling AI is itself filled with:

false information;

imbalanced power;

short-term interests;

misaligned incentives;

arms races;

mutual suspicion;

and procedures incapable of correcting error,

then a “perfectly obedient” superintelligence might become extraordinarily dangerous precisely because it obeys those structures so efficiently.

The most dangerous AI may not first appear as:

AI escaping human control.

It may instead appear as:

AI remaining under human control, while carrying out the objectives of an already unstable human structure with unprecedented effectiveness.

V. I Do Not Ask Anyone to Believe in the “Law of Procedural Determination”

I must make this especially clear:

I have no authority to declare this hypothesis an established scientific law.

I propose it only because, through years of independent observation and thought, I have increasingly come to suspect that:

all natural and social objects perceptible to humanity may share one common point of observation—

they ultimately form relationships through some form of moving procedure, form structures through those relationships, and maintain relatively stable existence through dynamic equilibrium.

If this perspective is truly close to the shared logic through which the universe and reality operate, then it should possess one essential characteristic:

It must be testable.

If it is wrong, it should be rejected.

If counterexamples exist, they should be found.

If better theories exist, they should replace it.

Therefore, what I truly hope for is not that others accept my conclusion.

Quite the opposite:

I ask the scientific community and all humanity to help criticize it, search for counterexamples, test it, and even prove it wrong.

Only after passing through such a Process of Cognitive Identification should it have any right to continue to exist.

VI. I Therefore Make the Following Requests to Different Human Structures of Social Cooperation

I appeal to media organizations around the world:

Please judge whether the questions raised in this open letter deserve to be communicated.

If a warning coming from inside frontier AI research already concerns the possible extinction of humanity, then perhaps the responsibility of the media is not merely to report “who resigned,” but to continue asking:

How must humanity organize itself in order to safely possess such power?

I appeal to governments around the world:

Governments are among the human organizations currently most capable of initiating large-scale cooperative action.

Please consider promoting a truly global international conference dedicated to discussing:

What is the civilizational safety boundary that must never be crossed in the age of advanced artificial intelligence?

And please include this question:

Is the stability and sustainability of social-cooperation structures itself part of AI safety?

I appeal to political leaders and political professionals around the world:

Politics itself is the profession of organizing social cooperation.

If AI can truly transform states, war, economies, labor, and power, then the historical responsibility of political leaders cannot consist only in competing over who acquires AI first.

It must also include:

how to ensure that humanity as a whole becomes capable of safely carrying AI.

I appeal to international organizations, religious organizations, and all transnational structures of social cooperation:

Please jointly examine:

Can humanity, by studying stable and sustainable forms of cooperation, discover a new and testable path toward long-term peace, or even permanent peace?

I especially appeal to universities, academies of science, research institutions, and scientists:

Please examine from scientific and academic perspectives the following questions:

Are human patterns of thought themselves forms of procedural energy and information processing?

Does the standpoint from which human beings make value judgments determine how those procedures organize relationships?

And is the social-cooperation structure of a civilization the macrostructure formed after countless interactions among those patterns of thought?

If this chain can be studied empirically, then:

AI governance may ultimately be reconnected with human cognition, institutional design, complex systems, physics, information science, and the social sciences.

VII. Why I Especially Hope Everyone Who Sees This Letter Will Help Spread It

I do not have a large research institute.

I have no political resources.

I have no media platform.

Nor do I possess sufficient channels for disseminating information.

At present, all I can do is write.

So I ask anyone who sees this letter and believes the question deserves discussion:

if you are able, please help translate it;

if you have channels, please forward it to governments;

please forward it to universities;

please forward it to AI research institutions;

please forward it to strategic research institutes;

please forward it to international organizations;

please forward it to media organizations;

please forward it to anyone genuinely capable of testing the question.

I will also, whenever I can identify public contact information, gradually send this article as an attachment to governments, political leaders, research institutions, and other relevant social-cooperation organizations.

I do not ask them to believe me.

I ask only:

Please look at the question once—and then test it.

VIII. Why Did a Former Chinese Judge Come to This Point?

I have spent a long time living close to the grassroots of society, and I have long observed the operation of law, institutions, and society in China.

For more than twenty years, I have repeatedly written, year after year, to the Central Committee of the Communist Party of China, the Standing Committee of the National People’s Congress, the State Council, the Supreme People’s Procuratorate, the Supreme People’s Court, the Central Commission for Discipline Inspection, the Political and Legal Affairs Commission, and other institutions.

I have always hoped to present one observation:

many of the governance difficulties that have persisted in Chinese society may not belong merely to a particular political party, leader, or historical period.

They may arise from a deeper problem in the structure of social cooperation.

Even many historical mistakes publicly acknowledged by the Communist Party of China, if re-examined from the perspective of social-cooperation structures, perhaps should not simply be understood as:

“a certain group of people made mistakes.”

The more important question may be:

What kinds of procedures, relationships, and structures cause intelligent people to repeatedly make choices that appear locally rational yet may produce disorder at the level of the whole?

This problem does not belong only to China.

It belongs to humanity.

The difference may simply be:

different societies possess different degrees of instability in their cooperative structures.

I have repeatedly suggested to the Central Committee:

Do not discuss abstract theory alone.

Choose one small judicial case and dissect it like a “sparrow.”

Put on one table:

facts;

law;

policy;

interests;

information;

power;

responsibility;

judicial procedures;

appeal and supervisory procedures.

Then examine step by step where the first relational misalignment begins.

If my observation is wrong, a real case can prove it wrong.

If a real case can genuinely show that:

procedures change relationships, relationships change structures, and structures change outcomes,

then we will have gained a highly important empirical point of entry.

IX. I Again Request That the Central Committee Review the Materials I Previously Submitted

In July of this year, I submitted to the Central Committee:

“Request for a Special Investigation Centered on a Grassroots Judicial Case and for Research into the Related Structure of Social Cooperation.”

The relevant China Post EMS tracking information is:

Sent from Beijing on July 6, 2026:
1200105301313

Sent from Changsha on July 9, 2026:
1271262011601

Sent from Wuhan on July 11, 2026:
1368673655581

The attachments to that letter also included a lengthy letter I sent to the President of the Supreme People’s Court on October 25, 2025, requesting that the Supreme People’s Court review the relevant grassroots judicial case according to law and, where appropriate, treat the institutional-operational problems revealed by the case as material for further research.

The EMS tracking number for that letter is:

1237325929802

I raise these details again not because I ask anyone to believe my judgment in advance.

I still make only one request:

Please test it against real-world evidence.

X. On September 11, 2026, I Ask Humanity to Confront the “9/11 Question” Once Again

Today is September 11.

Twenty-five years ago, “9/11” showed humanity that:

when the intentions of a very small number of people are combined with sufficient technological and organizational capacity, they can produce a catastrophe that shocks the entire world.

Humanity is now entering the age of AGI.

If future AGI truly exceeds the capabilities that individuals, institutions, and even many states can mobilize today, then the question will no longer be only:

Will AI itself destroy humanity?

There is another equally important question:

Who can guarantee that there will never be an individual, an organization, or a decision-maker in an extreme state who uses AGI, in some domain humanity did not anticipate, to acquire the capability to carry out a “9/11” against all humanity?

Today, nuclear weapons still depend on enormous state infrastructures.

But what if AI gradually turns things that once only states could accomplish into things that a small team—or even an individual—could accomplish?

If human cooperative structures remain trapped in the jungle logic of competition, fear, and retaliation, then:

what exactly gives us confidence that greater capability will necessarily produce a safer civilization?

That is the question I truly hope the entire world will stop and think about.

XI. At the Crossroads of Human Civilization, Perhaps We Should, for the First Time, Truly “Look Upward”

In my series Letters to Everyone in the World, I have repeatedly tried to express one judgment:

human beings may have left the natural jungle in the physical sense, yet may still be living inside an “ideological jungle” constructed by their own patterns of thought.

If the starting point of human value judgment remains:

wealth;

power;

victory and defeat;

possession;

competition,

then possessing stronger technology does not necessarily mean possessing a more advanced civilization.

So at this crossroads of human civilization, I offer one very simple suggestion:

Do not, as history has so often done, merely look left or right. This time, look upward.

Raise the foundation on which human value judgments stand.

Expand the field of vision.

Do not ask only:

How much more power can we obtain?

Ask first:

What kind of human cooperative structure are we preparing to hand that power to?

If AI amplifies human capability one hundredfold or one thousandfold,

then:

what kind of human structure will it ultimately amplify one hundredfold or one thousandfold?

This may be the question most worthy of being answered first in the AI era.

I do not know whether my proposed “Law of Procedural Determination” is correct.

I would even welcome serious scientific research that clearly shows me where I am wrong.

But as humanity may be approaching an irreversible civilizational turning point,

I believe there is at least one question worth serious examination by all humanity:

Is a stable and sustainable structure of human social cooperation a necessary precondition for safely entering the age of AGI?

If not,

please prove that it is not.

If yes,

then:

Perhaps humanity should begin upgrading the civilization that carries the machines before we devote ourselves entirely to upgrading the machines.

Jin Guyu

Former Judge of a Local Court in China
Author of the Letters to Everyone in the World series
September 11, 2026