TRANTOR LABS was founded in Singapore in April 2023. We are a Singapore-native, philosophy-first research lab studying human existential risk in the age of AGI.
For more than two years, we have made few systematic public releases and have done little public promotion around the lab itself. To the outside world, TRANTOR LABS may have seemed quiet, or even unfamiliar. But that quiet did not mean the work had not begun. On the contrary, the past two years have been a period of internal research, engineering reflection, philosophical clarification, conceptual iteration, and systematic consolidation.
Today, we are beginning to release this work publicly.
TRANTOR LABS does not begin from a single technical problem. We study what happens when advanced AI systems enter the structures through which human beings act, know, govern, judge, and make meaning. More specifically, we ask whether humanity can continue to survive, know reality, govern together, preserve agency, and sustain meaning as advanced AI becomes embedded in human civilization.
This question is closely connected to AI safety, alignment, governance, and AI assurance, but it also goes beyond the usual boundaries of technical safety. Advanced AI is not merely a more powerful software tool. It is gradually becoming an actor, a knowledge interface, and an institutional mediator. It may influence how humans act, how humans understand reality, how institutions make decisions, and how individuals form judgment. It may increase productivity, but it may also reshape the conditions under which humans understand value, dignity, and meaning.
The core question of TRANTOR LABS can therefore be stated as follows:
What conditions must hold for humanity to continue as a civilizational subject after AGI or advanced AI systems enter the world?
Why We Have Been Quiet
Over the past two years, TRANTOR LABS did not rush to publish unstable concepts or prematurely turn itself into a public narrative. The reason is simple: the questions we encountered required a long period of internal digestion and repeated calibration.
In the study of advanced AI systems, many questions appear at first to be engineering problems. But when examined more deeply, they move into the level of foundational categories. How should a long-term AI system handle memory? At what point does an AI system move from answering questions to participating in action? How can an AI-generated action path be audited? When a human clicks confirm, has human judgment actually participated in the formation of the decision? Is explanation the same as audit? Where does responsibility reside? How is reality anchored? How can human agency continue to hold in an AI-mediated environment?
These questions cannot be answered through quick positioning statements. They require continuous calibration across engineering experience, philosophical analysis, system design, and governance contexts. Many early judgments had to be revised. Many concepts had to be renamed. Many boundaries had to be tested again and again. The work of the past two years has been to allow these questions to move from scattered concerns into a structure that can be publicly discussed, continuously studied, and released over time.
This is why we are beginning the public release now. This is not the beginning of the research. It is the beginning of its public expression.
Where We Began
TRANTOR LABS began from the construction and reflection of long-term AI systems.
In our early work, we encountered a series of concrete questions around AI companionship, long-term memory, identity continuity, human-AI relationships, action recommendation, responsibility boundaries, and audit mechanisms. At first, these questions looked like questions of product design, system architecture, or human-computer interaction. But as the research developed, we came to see that they were not ordinary product details. They were foundational category problems that advanced AI systems must face before entering human life, organizations, and institutions.
If we do not know to whom memory belongs, long-term memory systems cannot truly converge. If we do not know who is acting, agentic systems cannot truly converge. If we do not know where responsibility resides, governance structures cannot truly converge. If we do not know whether explanation is the same as audit, safety systems cannot truly converge. If we do not know whether human-in-the-loop still means human judgment-in-the-loop, human-AI collaboration cannot truly converge.
These questions led us to a basic judgment: the place where advanced AI engineering often fails to converge is not code, but categories.
Identity, memory, action, responsibility, audit, reality, agency, and meaning are not secondary topics to be discussed after a system is complete. They are foundational objects that must be defined and tested before advanced AI systems can responsibly enter the world.
This is how TRANTOR LABS gradually formed its method:
Philosophy defines the problem. Engineering tests the answer.
Philosophy clarifies objects, boundaries, and evidential conditions. Engineering tests whether those objects, boundaries, and evidential conditions can actually enter system architecture, runtime processes, audit mechanisms, and real-world feedback.
From Long-term AI Systems to Human Existential Risk
At first, the question was whether an advanced AI system could be trusted, entrusted, and audited over time.
But as the research developed, the question expanded.
When AI mainly functions as an answering system, it is natural to focus on whether outputs are safe, whether content is accurate, and whether responses are compliant. But when AI begins to take on tasks, call tools, enter organizational workflows, influence real-world action, and become a knowledge interface through which more and more people understand the world, the problem is no longer located only at the level of output.
We must ask how action is formed, how evidence is selected, how options are ranked, how risk is assessed, and whether human judgment participates in the formation of decisions or merely confirms them at the end.
This is how TRANTOR LABS gradually moved from long-term AI systems, AI safety, alignment, governance, and structural safety evidence toward a higher-level question:
In the age of AGI, can humanity continue as a civilizational subject?
Under this question, human existential risk should not be understood only as the possibility of biological extinction. It also includes the possibility that the conditions under which humans continue as agents of judgment, responsibility, reality-seeking, collective governance, and meaning-making may be irreversibly, or very difficult to reversibly, weakened.
TRANTOR LABS initially organizes this research direction around five dimensions: survival, reality, institutions, agency, and meaning. Future papers and essays will develop these dimensions in greater depth. Here, we only want to state why they arise: once AI deeply enters action, knowledge, institutions, judgment, and meaning, the risk is no longer only whether systems fail. It is also whether human beings retain the capacity to understand the world, participate in action, bear responsibility, govern together, and create meaning.
How We Will Release Our Research
Over the coming year, TRANTOR LABS will gradually release more than twenty papers and long-form studies. These are not hastily produced works, nor are they attempts to manufacture attention by chasing current trends. They arise from more than two years of sustained research, repeated iteration, and systematic consolidation around long-term AI systems, structural safety evidence, AI governance, human agency, civilizational risk, and action philosophy.
The first stage will begin with three general papers on civilizational risk. These papers will establish the first layer of problem definition. They will address a set of foundational questions: Who is at risk? What are the primary risks? How do these risks emerge, spread, and reshape the conditions under which humans act, know, govern, judge, and make meaning?
After these three general papers, we will continue to release more than twenty specialized papers and long-form studies on AI safety, AI governance, AI alignment, AI risk, structural safety evidence, reality risk, institutional risk, agency risk, meaning risk, personal cognitive OS, and the conditions for long-term coexistence between humans and advanced AI.
The publication rhythm of the coming year does not mean that the research is just beginning. More accurately, it marks the systematic public release of work that has been internally developed and consolidated over the past two years. Our aim is to allow these concepts, frameworks, and problem definitions to enter public discussion and to provide more stable theoretical anchors for future work in governance, education, standards, engineering, and public thought.
Why We Begin with Papers
TRANTOR LABS begins its public release with papers not because we believe papers can solve everything.
Papers cannot replace public discussion. They cannot replace institution-building. They cannot replace education, engineering, governance, or community. They do not automatically change the world.
But papers can provide long-term anchors for concepts.
Many important questions about AI are now moving rapidly through social media, conferences, policy documents, product narratives, and industry reports. This speed has value, but it can also make concepts vague, unstable, and short-lived. If a concept has no stable definition, no citable text, no retrievable version, and no argument that can be reviewed, it is difficult for that concept to enter longer-term governance, education, standards, and institutional memory.
This is why we treat papers as DOI-based scholarly anchors.
They are not the end point of communication. They are anchors for communication. They allow future blog posts, Substack essays, Research Center objects, public discussions, policy responses, educational materials, and institutional translation to return to a stable source.
Papers cannot replace public discussion. But without papers, ideas too easily drift.
Beyond Papers
For research to enter the world, TRANTOR LABS will gradually build an infrastructure for research publication and public explanation.
The website will serve as the official research entry point for public statements, paper releases, stage summaries, and official essays. The Research Center will organize papers, concepts, figures, reading paths, version records, and external discussions. Its goal is not to become a generic resource library, but to become a system of canonical research objects, so that each paper, each core concept, and each important discussion can return to a clear research object.
The Blog will serve as a layer of research explanation. It will publish Official Lab Notes and Official Lab Stories in the voice of the lab. Official Lab Notes will be used for paper releases, research roadmaps, and stage summaries. Official Lab Stories will explain how TRANTOR LABS moved from long-term AI systems to philosophy-first AI safety, structural safety evidence, and human existential risk research.
KunYuan’s Substack will serve a different function. It is not a replacement for the website, nor is it an archive for papers. It will be a first-person space for theoretical teardown, thought transmission, and a high-density reader community. The website will say: this is a research system that can be cited, verified, and preserved. The Substack will speak in the first person and bring these ideas to real readers.
In the first stage, we will proceed carefully: publish the papers, establish the basic objects in the Research Center, explain the research through the Blog, and bring the key ideas into contact with real readers through KunYuan’s Substack.
What Comes Next
TRANTOR LABS will first release the first general paper on civilizational risk. It asks a foundational question: in AI-related human existential risk, who is the risk-bearing subject?
This paper is not a complete risk map, and it is not a governance proposal. Its task is to provide the first layer of ontological coordinates for the research that follows. Only after clarifying who is at risk can we continue to discuss what the risk is, how it occurs, and why we have misread it.
After the paper is released, we will publish an official note on the website, create the corresponding Paper page in the Research Center, continue explaining the research background through the Blog, and publish more reader-facing theoretical essays through KunYuan’s Substack. The seven general papers on civilizational risk will then be released over time, followed by more specialized papers and long-form studies throughout the coming year.
TRANTOR LABS was founded in Singapore in April 2023. For more than two years, we have been researching, iterating, and consolidating internally. Now this work is beginning to enter public space.
Our long-term commitment is not to make AI more powerful. It is to ask what must hold true, when powerful AI enters the world, so that humanity does not lose the conditions under which it can continue to survive, know reality, govern together, preserve agency, and sustain meaning.
