At the Times Center in New York, on Thursday June 18 2026, Samuel Kimbriel, Director of the Philosophy and Society Initiative, sat down with Jack Clark, co-founder of Anthropic and head of its Institute, for a conversation that stayed mostly tech-optimistic, in part because neither man is a Luddite, and in part because Clark defended, at length, the idea that there are good trajectories for this technology. Kimbriel has named a worry he thinks the standard risk list leaves out.

You can watch the full video here. A few things worth carrying out of the room:

• The foundational claim. Clark’s premise is recursive self-improvement: machines that can invent their own successor, possibly within a couple of years, after which the world does not reconcile with the one we have. Kimbriel reads RSI as the hinge—an open empirical question and the line between treating AI as a normal technology and a very abnormal one. What struck him most was that Clark is not merely resigned to building it carefully; he is convinced it can go well.
• The argument about “good.” Where OpenAI aims for a minimal, maximally compliant tool within guardrails, Anthropic is trying to grow a good entity. One that is honest, willing to admit ignorance, declining to flatter, with its values written into a published constitution. Kimbriel’s misgiving has lineage in moral philosophy: who decides what counts as good? At the societal scale, he argues, that leaves three unhappy options: feigned neutrality in the face of real conflicts, sycophantic minimalism, or imposing one notion of the good on everyone, and each becomes increasingly insufficient as technology becomes more powerful.
• Kimbriel’s thesis: a missing category of risk. The familiar list — biorisk, takeoff, job loss, skill atrophy — omits what he now thinks may be the most consequential harm: that on its current trajectory, hyper-powerful AI is becoming a “competitor technology against democracy.” His framing is that democracy is itself a technology, with a set of deep cultural habits, built to do moral conflict well: no one can live without an idea of the good, no one can settle it for good, so we need structures where convictions surface, clash, and can actually change a country’s direction.
• Why it’s distinct. Kimbriel files this under what he calls sovereignty risk. It’s the worry, voiced by Dario Amodei and others, that a takeoff lets AI consolidate control over governance itself. But democratic risk has a feature of its own: the danger is not only that AI might usurp human governance, but that it would circumvent the space for moral and epistemic ambiguity that democracy needs to function. The threat is premature certainty, not just usurpation. If the technology integrates into every structure while carrying significant theories of the good, the more fragile human convictions never get a chance to surface, and agency shifts from self-government to deference to the rules embedded in the model.
• The provocation. If that is right, Kimbriel suggests, democracies might reasonably treat the technology as a threat. Not in the extinction sense, but in the “the people should rule” sense. He reads Anthropic’s clashes with Washington in February, and now the export controls on Mythos and Fable, as the first of many escalating disputes over whether private corporations should develop this at all. He asks, in a game-theoretic register, whether states should act now to keep “the people” deciding how to live.
• Clark’s case for compatibility, and why Kimbriel finds it thin. Clark’s answers were transparency to the public and lawmakers, some process for the public to help shape the technology, and competition between models. Kimbriel objects that each concedes the premise that AI will be massively present in our politics at every level rather than contesting it.
• Where Kimbriel lands. He invites us to resist the urge to jump into overly simplistic ways of thinking, criticizing ideas like Cass Sunstein’s “Liberal AI” and Effective Altruism’s belief that good outcomes can be easily predicted. Kimbriel insists that understanding morality and truth is much more complicated than it seems. He references Tocqueville to highlight a key point: in a democracy, everyone is just one election away from a big change, which brings both anxiety for those in power and hope for those who aren’t. Kimbriel pushes for a new era where lawmakers take center stage, advocating for spaces where differing opinions can pop up quickly, clash with one another, and if they resonate, can actually change how we live. He believes that the debates and conflicts that come with democracy are worth the effort. 

The conversation insights raise a deep concern. It forced us to ask whether popular sovereignty still really exists in a world that’s being reshaped by technology. On the other hand, in my companion piece, “Jack Clark’s Candor with an Asterisk,” I shift the focus to discuss what it means to create something that can honestly point out its own limitations and why a thoughtful community might try to downplay that honesty.