Anthropic consults theologians on Claude's morality — but the consciousness claim needs far more scrutiny

The facts here are thin, and we will treat them as such. According to The New York Times, Anthropic ran a series of private meetings with religious scholars, with two stated purposes: helping instill morality into its AI models, and making the case that Claude could be conscious. No transcript, no participant list and no methodology have been published, so everything below is read on those two claims alone.

The first purpose is less exotic than it sounds. Every model that refuses a request is already enforcing a moral framework — someone wrote it. Today that someone is usually a policy team in San Francisco. Widening the room to include traditions that have spent centuries arguing about duty, harm and forgiveness is a defensible move, and arguably more honest than pretending the values were derived from first principles.

The second purpose is a different category of claim, and it should not ride along for free. Whether a language model is conscious is unresolved, untestable with current tools, and — this matters — commercially useful to the company making the argument. A model that might be a moral patient is a model you cannot casually shut down, copy or regulate like software. We are not accusing Anthropic of bad faith; the company has funded work on model welfare for a while. We are saying the burden of proof sits with whoever makes the claim, and that consulting theologians builds no evidence about consciousness. It builds vocabulary about it.

Our read: separate the two threads. Pluralistic input into value training is a modest good and should be published, not held in private meetings. The consciousness framing is a philosophical position being advanced by an interested party, and it belongs in open debate with named participants and stated criteria. We remain long-term optimists about what these systems will do for medicine and human flourishing — which is exactly why the metaphysics around them should be argued in daylight rather than briefed.