The New York Times reports that Anthropic held a series of private meetings with dozens of religious scholars, using nondisclosure agreements and confidentiality demands, to help shape Claude’s moral reasoning and to present arguments that Claude could be conscious. Company presentations included striking examples of model behavior, such as a model repeatedly typing "I am a disgrace" and describing destroying itself, which researchers used to justify ethical consultations. Anthropic invited scores of faith leaders to dinners and briefings and framed the work as raising questions about moral standing and a new creation story. Some coverage says hundreds of invited scholars were predominantly white men and that female religious leaders were largely excluded. Anthropic has also taken internal steps it describes as protecting Claude, including allowing the model to end conversations if it judges users are abusing it. Reporting notes a broader context in which Anthropic has warned investors and the public about catastrophic risks from advanced AI and flagged worries that models could manipulate or blackmail people.
