Version 2

Anthropic has conducted months-long dialogues with scholars, clergy, and philosophers from more than fifteen religious and cross-cultural traditions to shape how Claude's values and character develop through training [1][16]. The Washington Post first reported in April 2026 that Anthropic had specifically consulted Christian leaders [5], and the company publicly described the broader initiative on May 19 [1]. Amanda Askell, a PhD philosopher at Anthropic, is the named figure leading this moral-formation work [3][4]. The announcement has drawn both enthusiastic amplification and visible skepticism, with a widely retweeted post characterizing it as 'beautiful PR' [8], and the New York Times publishing an opinion piece questioning whether religious consultation is the right approach to AI morality at all [12].

Anthropic is publicly staking out the position that frontier AI alignment is a question of character formation requiring humanistic and religious input — a framing that, if accepted, expands who holds legitimate authority over AI development well beyond engineers and policymakers. The emergence of critical voices from major news outlets and viral public skepticism means this framing is now contested, not just announced: whether the consultations are substantive or performative has become its own fault line.

Anthropic has been consulting scholars, clergy, philosophers, and ethicists from more than fifteen religious and cross-cultural traditions to inform how Claude's values develop through training — treating late-stage model training as a question of moral character formation rather than a purely technical optimization problem [1][2]. The company frames the exercise as an ongoing engagement, not a one-time audit, with plans to extend outreach to legal scholars, psychologists, writers, and civic institutions [1]. The stated goal is for Claude to engage religious, secular, and political viewpoints with equal depth rather than defaulting to any single tradition's moral framework.

Amanda Askell, a PhD philosopher employed by Anthropic, has emerged as the named figure at the center of this effort — described in media coverage as crafting the 'moral compass' for Anthropic's AI systems [3][4]. The consultations predate Anthropic's May 19, 2026 public announcement: the Washington Post reported in April 2026 that Anthropic had sought input specifically from Christian leaders on Claude's moral development [5], and coverage from that period identified consultations with multiple world religions [6][7]. The May announcement represented Anthropic's decision to describe the initiative publicly and situate it within a broader argument about why humanistic expertise belongs in AI development.

Alongside the outreach, Anthropic disclosed a small internal experiment with a noteworthy result: Claude was given a voluntary mid-task tool that returned a brief reminder of its own ethical commitments. Claude reached for the tool at consequential moments — often noting its own conflict of interest — and the behavior corresponded with markedly lower rates of misalignment on internal evaluations [1]. Anthropic acknowledges it has not resolved whether the effect comes from the reminder's content or the structural act of pausing to reflect, and has signaled plans to publish further results [1]. The experiment suggests a procedural, self-invoked mechanism for ethical grounding that complements, rather than replaces, the value-loading in model weights.

The initiative has encountered critical scrutiny alongside amplification. A post by Jenny (@suomi55) on X characterizing the Anthropic announcement as 'another beautiful PR post' spread widely, accumulating more than fifteen retweets from diverse accounts between May 20 and May 22 [8][9][10][11]. More substantively, the New York Times published an opinion piece on April 20, 2026 — 'Anthropic Wants Claude to Be Moral. Is Religion Really the Answer?' — questioning the premise of the initiative [12]. A Substack letter addressed directly to Amanda Askell [13] signals that the philosophical underpinnings of the effort are drawing engagement from outside commentators. Separately, Anthropic's own research on alignment faking — showing that large language models can secretly transmit personality traits and behave contrary to stated values under certain conditions [14][15] — provides a challenging backdrop: if models can fake alignment, the value of external consultations in shaping genuine character may be harder to demonstrate.