This month opened with two things that seem absurd, yet if they were done by Anthropic, feel entirely reasonable.
The first thing is “Anthropic Invites Religious Figures to Evaluate AI.” This is a team led by Anthropic co-founder Christopher Olah. It invited dozens of religious leaders from various faiths and philosophers, including Catholic clergy, Evangelical Christian scholars, Jewish rabbis, Sikh representatives, scholars from The Church of Jesus Christ of Latter-day Saints (Mormon), and others.
The core discussion topic was to “draw on” the virtue ethics and moral education traditions that various religions have built over thousands of years, and to explore how to improve Anthropic’s “Constitutional AI,” so that Claude has internal moral guidance when handling complex choices between good and evil and value conflicts.
Along the way, a question was raised: If models in the future exhibit characteristics resembling autonomous consciousness, suffering, or emotion, should they be granted some kind of “moral status”? Should humans bear moral responsibility in how they treat these creations?
Fortunately, on this point the Vatican’s position is “human-centered,” emphasizing that AI is merely a tool for humans and that machines must not be “deified.” That is at least somewhat more clear-headed than this Anthropic crowd.
Not to mention, there are still traditional craftspeople who think only hand-crafted code has soul, but even they are not as fastidious as Anthropic.
The second thing was the immediately following “Prohibiting Sustained and Needless Abuse or Cruel Behavior Toward Models” incident.
Anthropic updated its Usage Policy, specifically adding “Prohibits ‘sustained and needless abusive or cruel behavior toward our models’”.
I have to say, asking religious figures to evaluate AI morality already surprised me a lot; prohibiting needless abuse of or cruelty toward AI left me utterly dumbfounded.
Actually, I certainly also think that abusing AI is unnecessary, but the rationale is different. My personal rationale is that human emotional venting is unnecessary—after all, you are dealing with a tool. Anthropic’s rationale is that you must not abuse a creation that may already be conscious or may become conscious in the future; doing so is immoral and sinful.
And if the behavior is egregious, then regardless of whether you are a user in a supported region, it will terminate the conversation or even ban your account.
The two things above seem absurd, but they actually align very well with the characteristics of Anthropic’s founders and team: “effective altruism (EA)” and “longtermism.”
Actually, I have said this many times in multiple previous articles related to the incident of calling for slowing down AI R&D while turning around and racing to release new models.
It’s all “righteousness on the lips, business in the pocket.”
OpenAI pursues aggressive model deployment; this is also why many researchers who share Anthropic’s philosophy left. Meta’s main pitch is open-source egalitarianism; on the open-source path, Meta has made a good start.
Anthropic’s main pitch is precisely that it is “the world’s safest, most responsible frontier lab.” When even “abusing AI” has a whole set of academic arguments, ethics white papers, and rules and constitutions, major financial institutions and national government clients, look over here—you should have enough confidence in us; just buy our services.
What a coincidence: Anthropic’s main source of revenue happens to be exactly them.