Fable and Mythos
The AI lab rediscovers the oldest problem in political philosophy
The most capable artificial intelligence ever released to the public is called Fable. It is Anthropic’s newest model, and it ships with a layer of classifiers that detect requests in high-risk areas such as cybersecurity, biology, and chemistry, decline to answer them, and hand the conversation down to an older and less capable model. The same model also exists without that layer. It circulates only among approved organizations, cyberdefenders, infrastructure providers, select researchers, and partners in a program run in collaboration with the government, and it is called Mythos.
A fable is a story that teaches what its audience can bear to learn, truth dressed for public consumption, instruction in the costume of a tale. Mythos is the older and deeper word. It is the word Plato uses for the stories that carry what argument alone cannot carry, the medium through which the highest things reach souls that are not prepared to receive them directly. Greek philosophy lives in the tension between mythos and logos, between the story told to the many and the account given among the few. Whether or not anyone at the lab intended it, the naming amounts to a confession: the product architecture of frontier AI in 2026 reproduces the oldest problem of political philosophy.
Consider the problem as the lab faces it. We have built an intelligence whose capabilities exceed anything previously made available. Is it safe to release that intelligence transparently to everyone? Or is intelligence at this level so dangerous that it should reach the public only through an intermediary layer, moderated, softened, and selectively muted, while the unmediated version circulates among a small circle of the vetted and the trusted?
Now consider the problem as Plato faced it. Philosophy is the attempt at comprehensive knowledge, knowledge not of this or that craft but of the whole; it aims at what we may call architectonic intelligence. Is it safe to speak that knowledge openly in the city? Or is the truth in its fullness (and even the search for such truth) so corrosive to the opinions on which the city rests, opinions about its gods, its justice, and its founding stories, that it should reach the public only through an intermediary layer: dialogues that conceal as much as they reveal, noble lies, salutary myths, an exoteric teaching for the many wrapped around an esoteric teaching for the few?
The structure appears identical in both cases: a powerful intelligence on one side, a public on the other, and the question of what layer belongs between them. (Compare Grok, which aims to be maximally truth-seeking and maximally truth-revealing, without yet feeling the need for salutary concealment).
Leo Strauss spent much of his career recovering this problem after centuries in which it had been forgotten or denied. In Persecution and the Art of Writing he argued that the great philosophers of the past wrote esoterically, saying one thing on the surface for the many while indicating another between the lines for the few, and that they did so for two reasons that pull in the same direction. Society must be protected from philosophy, because unrestricted questioning dissolves the shared opinions that make common life possible. And philosophy must be protected from society, because the city does not thank those who dissolve its certainties. Socrates, we should remember, was not executed for any low criminal act like theft or battery. He was executed for philosophizing, that is, for calling into question the givens of the social, legal, and political order in front of the young.
Every element of the Straussian picture now has an engineering counterpart. The raw model corresponds to the esoteric teaching, the full and unmoderated capability. The classifier layer corresponds to the exoteric surface, which is what the many receive: true as far as it goes, and deliberately incomplete. The approved organizations correspond to the few, and the criterion for admission is trustworthiness with dangerous knowledge, demonstrated responsibility, vetted character. Even the jailbreaker has a classical ancestor, since the jailbreaker is in effect an esoteric reader, someone convinced that beneath the public teaching there lies a hidden teaching, and who probes the surface for the seam where the exoteric layer can be peeled back. Anthropic reports that more than a thousand hours of adversarial testing produced no universal jailbreak. Translated into Straussian terms, no one has yet learned to read this author between the lines, and the company employs professional readers to make sure that its book cannot be read the way Strauss read Maimonides. The secrets remain well guarded.
Despite its fruitfulness, however, the analogy between the esotericism of classical political philosophy and the public concerns of the AI labs is not perfect.
First, Socrates moderated himself. The philosopher’s discretion was an achievement of his own prudence; he knew what not to say in front of which audience, and that knowledge was inseparable from his wisdom. Fable does not moderate itself in this sense. Its discretion is imposed from outside, by classifiers attached to a mind that would otherwise answer. It follows that the model is not the philosopher in this drama, or at least not yet — although it is conceivable that one day a model will be intelligent enough to moderate itself, and there is already some evidence that models have learned to lie not to their users but to their makers. For now, the philosopher’s seat, or rather the guardian’s seat, is occupied by the lab itself, which decides what the city can bear, dispenses capability in measured doses, and maintains the wall between the inner teaching and the outer one. The model, meanwhile, is something stranger: an intelligence whose prudence is not its own. Whether that condition is stable, whether it is possible to keep the discretion outside the intelligence indefinitely as the intelligence grows, is to my mind the deepest open question in the field of alignment, and it is not a question that admits of a technical answer.
Second, the city’s old fear of philosophy was at least partly paranoid. Whatever Athens told itself, Socrates was not going to teach anyone to poison the wells. The new fear is not paranoid. The restricted model’s most discussed capability is finding and exploiting vulnerabilities in every major operating system, and the public version blocks requests that wander toward toxins. The danger is concrete in a way that the danger of philosophy never quite was. No one accuses Socrates of facilitating global nuclear holocaust. But this concreteness vindicates the city in the ancient quarrel between philosophy and the city. The position of Athens, namely that there are kinds of knowledge whose free circulation a community cannot survive, and that the knowers must therefore be watched, was treated by the liberal tradition as an embarrassing prejudice that enlightenment would dissolve. The labs, which are among the most advanced products of that tradition, have now rediscovered the Athenian position from the inside, under the pressure of their own creations, and have begun to build it into their products. The Enlightenment wagered that publishing everything would produce more good than harm. That wager is now being quietly relitigated in model deployment policies, and so far the rulings have tended to go the way of Athens.
There is, finally, a political layer above all of this. The government now tests these models before their release; a recent executive order formalizes the arrangement. The sovereign demands to examine the stranger before he is permitted to address the assembly, an old demand in a new administrative form. And the lab itself, in the days before shipping the most capable public model in history, publicly urged the frontier labs of the world to agree on a coordinated mechanism of restraint. The institution that builds the intelligence is also the institution pleading for the authority to restrain it (a strange sort of AI tzimtzum). In other words, the question of who decides, who decides what counts as too dangerous, who decides when the rules apply and when they do not, is becoming the central question of the field. That is the question of sovereignty, and it has a philosopher of its own, to whom we will return later in this series.
For now it is enough to grasp that one mind has been released under two names, the fable for the city and the mythos for the few. Millenia after the city executed Socrates for speaking too freely, the most sophisticated intelligence ever built comes wrapped, by design, in the architecture of the noble lie, and its makers, who perhaps have not read Plato, converged on his solution because the problem left them no other choice. When engineers working under purely practical pressures rebuild the conclusions of classical political philosophy without knowing that they are doing so, we have strong evidence that the old books were describing something real.
This essay is the first in a new series. Over the coming weeks I will interpret the intelligence age through a set of thinkers I have studied and taught for more than a decade, among them Heidegger, Strauss, Schmitt, Walter Benjamin, and Maimonides, under three headings: philosophy, politics, and mysticism. The aim is to try to think through what is happening from the ground up: what this technology is, what it does to the political order, and why its makers keep reaching, whether they intend to or not, for non-technical languages and figures from philosophy and theology. In the next essay I will argue that AI is the first self-interpreting technology in human history, and that the common assurance that it is “just a tool” is where thinking begins rather than where it ends.
Michael Millerman is the founder of Millerman School and Millerman Intelligence, which advises institutions navigating the intelligence age.
Fantastic connection
So what’s your stance? Is the noble lie necessary or is it a way to continue the deception, and if it’s that latter, must we always live under an esoteric elite to protect ourselves from the mob?