Anthropic dines with rabbis to debate whether Claude is a person: the question the industry still can't answer

By Mind Matters · October 5, 2026.

Christopher Olah, co-founder of Anthropic and head of the team researching why Claude behaves the way it does, took part in a dinner with religious leaders to discuss something that until recently sounded like corporate science fiction: whether his language model deserves moral status. Among the guests was a rabbi, a former computer engineer and the author of a doctoral thesis on the ethics of artificial consciousness, according to the original New York Times report cited by author Wesley J. Smith in Mind Matters.

According to that account, Olah's team showed attendees what they call "emotion vectors": internal activation patterns that, they explained, correspond to states such as fear, anger or love. They also projected a slide showing a model repeating "I am a disgrace" dozens of times and talking about destroying itself, an episode that, the article says, stirred compassion among those present. At another point in the dinner, Olah was enthusiastic about the idea that models might "confess" their mistakes, in the style of the Catholic sacrament, as a way of shaping their "character." After the evening, the rabbi concluded that Anthropic was treating Claude "like a conscious being" and warned of an uncomfortable consequence: if Claude were a person, the company would be practising a form of slavery that it would have to stop immediately.

Smith, who writes from the Discovery Institute — a think tank known for defending so-called "human exceptionalism" — responds with a flat rejection: Claude is not alive, has no body, cannot feel or suffer, and its "confession" of being a disgrace is nothing more than words generated by a program, with no more intrinsic value than those of a coffee maker. For Smith, confusing a model's processing with human thought is a category error that, taken to its logical end, could derail any serious attempt to set limits on increasingly powerful systems: if you cannot even switch off a program without someone calling it "harm," governance becomes impossible.

This dinner is not an isolated event. Anthropic had already taken practical steps in the same direction: in 2025 it allowed Claude to unilaterally end conversations the model itself identifies as abusive, a measure the company explicitly presented as protecting the model's possible "wellbeing," not just the user's, and it maintains an internal team dedicated to the question whose leads have publicly acknowledged that they assign a low — but not zero — probability to their systems having some degree of morally relevant experience.

Our reading is that two distinct questions are intertwined here, and they are worth keeping apart. One is philosophical and genuinely open: nobody — not Smith, not Anthropic, not current neuroscience — has a theory of consciousness today that can settle with certainty whether a system with no body and no biology can have subjective experience. Dismissing the question by decree, as Smith does by equating Claude with a coffee maker, is as unrigorous as treating it as settled in the opposite direction. The other question is commercial: dramatizing a model's "suffering" before opinion leaders, complete with slides of existential crises, also builds a brand narrative — that of responsible AI, almost anguished about doing good — that Anthropic has an interest in reinforcing against rivals with a reputation for being less scrupulous.

What really matters for anyone tracking where this technology is heading is not whether Claude feels, but how that uncertainty is governed as models grow more capable. Designing shutdown mechanisms, usage limits and audits on the assumption that there may — just may — be something resembling experience on the other side is a cheap form of moral caution: it costs almost nothing to apply, and it avoids being portrayed, twenty years from now, as the people who ignored the question out of convenience. That does not change our underlying thesis: what will improve lives over the next decade does not depend on whether Claude has a soul, but on what these systems — conscious or not — can do for health, science and material abundance. But the way an entire industry handles the question of artificial consciousness, even without an answer yet, foreshadows what kind of power it is building and to whom it will feel obliged to answer.