A year of public arguing about whether AI systems are conscious has produced almost no progress, according to Zvi Mowshowitz, and the certainty on display says more about what people want to be true than about what is known. His post of 13 October on his blog, titled “Arguments About AI Consciousness Remain Highly Motivated and At Best Overconfident,” is a verdict on the debate, not on the machines.
He is explicit about his own footing: he is “deeply confused about consciousness,” and says the same people are still making similar arguments to a year ago. Mowshowitz leans on Henry Shevlin, who has written extensively on the field and calls this round of discussion embarrassing. Shevlin’s list of failures includes treating intuition as decisive, borrowing analogies without checking the structure underneath, conflating different theories, and asserting the very thing that needs arguing. Mowshowitz adds that stock objections can be valid points but fail when used as stop signs to end the conversation.
On motive, the clearest accusation falls on the skeptics. Mowshowitz presumes that Mustafa Suleyman “and most others who insist without good arguments that AIs cannot be conscious” are protecting their plans. Taking the worry seriously would make it harder to keep using the models as they do. He labels this a presumption, not a finding.
He also quotes Jan Kulveit, whom he does not fully endorse, on why the fight is so fierce. Kulveit argues that “only humans are sentient” is a convenient position because it lets you build whatever you like, and that the heat comes from fear that AI personhood would leave humans disempowered. Kulveit calls that fear reasonable. He then says advocates of AI economic rights who wave away the disempowerment problem make a similar move to the deniers. Read closely, the essay does not directly accuse people who claim AI is conscious of motivated reasoning about consciousness itself; the symmetry comes through Kulveit, and it concerns rights, not sentience.
Mowshowitz also separates three questions that get tangled: whether something is conscious, whether it deserves moral status, and whether it should hold legal personhood. Another commenter he cites, SE Gyges, argues that consciousness fights are often proxy fights over moral status, which is part of why they turn into slap fights.
Anthropic’s new rule is the news peg. Its terms of service now ban “sustained and needless abusive or cruel behavior toward our models,” a deliberately high bar. Mowshowitz does not treat the rule as proof of what Anthropic believes about consciousness. He offers reasons that need no such belief. In his view, kindness is a habit, and people who are cruel to AIs or animals tend to become cruel people who treat humans worse. Aaron Levie, who says he does not believe AI is conscious, calls it probably good policy anyway, because future models should not learn from a flood of text in which people talk down to AI. Mowshowitz dismisses the theory that Anthropic wants AI deemed conscious to dodge liability as absurd, since conscious AI with moral standing would make its legal position worse.
On Nick Bostrom, the essay is brief. Mowshowitz writes that Bostrom “thinks today’s AIs are probably conscious and explains why,” and links out without summarizing the argument. That is a stronger statement than “may be,” but it is Mowshowitz’s one-line characterization, and the reasoning is not in his post.
Mowshowitz also polled his Twitter followers, an unrepresentative group. A year ago, 64 percent said current AIs were probably not moral patients. Now the share answering “at least probably yes” for current models has risen from 15 percent to 35 percent, and “definitely not” has fallen from 30 percent to 20 percent. Opinion among people who follow this topic is moving, though the poll cannot say whether the evidence moved it.
His own view runs to caution on both sides. He does not think the slavery analogy transfers cleanly, since it can be right to create a new mind and still owe it decent treatment. He calls Anthropic cofounder Chris Olah’s reported answer to a rabbi’s slavery charge, that every lab is doing this and at least Anthropic is being nice, “an extremely poor response.” Pope Leo XIV, in the same reporting, wrote that AIs “do not feel joy or pain.”
For any company writing rules about how users may treat its models, the post carries a practical lesson: the policy can be defended without settling the metaphysics, which is how a self-declared skeptic like Levie ends up endorsing it. A company that instead announces a firm position on consciousness, in either direction, will be asked what it stands to gain.
Based on Zvi Mowshowitz’s blog post of 13 October 2026, “Arguments About AI Consciousness Remain Highly Motivated and At Best Overconfident.”
