AI theology: keep your gulliver, droog

On the record

Sam Altman’s October 3 X post objects to giving AI religious authority or surrendering human judgment to it. He calls that a “real safety issue.” Anthropic is not named.^[2]

In the interpretation department

Axios reads the remark as a veiled shot at Anthropic. That is its reading of the rivalry, rather than a target Altman identifies in the post. Axios says neither company immediately responded to its questions.^[1][2]

Already in the paperwork

Anthropic’s May 19 account describes discussions with scholars, clergy and others from more than 15 religious and cross-cultural groups. One resulting experiment gave Claude a tool that reminded it of its ethical commitments mid-task. The company reported less misaligned behavior on several internal evaluations, while acknowledging it had yet to separate the reminder’s effect from the pause to reflect.^[3]

Its January constitution announcement also expressly leaves model consciousness unresolved and says humans must remain able to oversee, correct and stop Claude. Those are stated intentions, not a certificate of reliable behavior.^[4]

The gulliver stays with its owner

By all means invite the philosophers. Then show the tests. If an ethical reminder helps, measure where it fails, too. Keep a hand on the controls, droogies; a machine that sounds wise has not earned the keys to your conscience.