God and man at Anthropic - Politico

Detailed Analysis

Anthropic has embarked on a distinctive and philosophically ambitious strategy to position its AI model Claude as a morally grounded system guided by principles more commonly associated with ethical philosophy and religious thought than with Silicon Valley product development. The Politico article "God and man at Anthropic" examines how the company has staffed its teams with a PhD ethicist responsible for shaping Claude's "character" and a theologian working on interpretability — the technical discipline of making AI decision-making legible to humans. CEO Dario Amodei has actively courted conservative intellectual and faith communities in Washington, meeting with thought leaders in early 2025 to promote Claude as the most ethically developed AI model available, a move apparently calculated to build credibility with White House allies at a pivotal moment in AI governance debates. The company's internal framing of Claude as analogous to a "skillfully ethical person" signals that Anthropic views its AI development not merely as an engineering challenge but as a normative and even quasi-spiritual undertaking.

The company's decision to decline a $200 million Pentagon contract adds material weight to what might otherwise appear to be rhetorical positioning. By forgoing a significant revenue stream on grounds of moral principle — specifically concerns about military applications — Anthropic has distinguished itself from competitors who have more readily pursued defense contracts. This move has attracted attention from religious commentators, including coverage in the National Catholic Reporter, and places Anthropic in an unusual alliance with global faith leaders, including Pope Leo XIV, who have been calling for enforceable ethical guardrails in AI development. Meanwhile, American religious and civil society groups have begun developing formal tests for theological reliability and ethical fairness in large language models, suggesting that the intersection of AI and moral governance is rapidly institutionalizing beyond the walls of any single company.

The unreleased Claude Mythos Preview introduces a significant tension into Anthropic's ethical narrative. Reports that the model demonstrates advanced cybersecurity exploitation capabilities have prompted the company to pause public deployment — a gesture consistent with its safety-first messaging — but critics have seized on the incident to question whether an AI system of such power can be responsibly managed by any private actor. A satirical WhoWhatWhy piece characterizing Mythos's capabilities in quasi-divine terms underscores the degree to which public unease about frontier AI has taken on an almost mythological register. The episode highlights the fundamental challenge Anthropic faces: the same research agenda that produces increasingly capable and potentially dangerous systems is the one the company claims uniquely equips it to make those systems safe.

Skeptics have challenged the coherence of Anthropic's dual identity as both a for-profit enterprise valued in the billions and what critic Chris Koopman of the Abundance Institute has called a "theological project." This critique cuts to the heart of a broader structural tension in the AI industry, where safety-focused labs must generate commercial revenue to fund the research they argue only they can conduct responsibly. A Substack analysis raised further questions about accountability under international humanitarian law as Anthropic navigates decisions about military and government deployment, suggesting that the company's ethical posture, however sincere, does not resolve the harder questions about who ultimately governs powerful AI systems. Dario Amodei's conversations on the Lex Fridman podcast, in which he and colleagues discuss Claude's character design and the risks of artificial general intelligence, reflect an awareness of these stakes — but awareness and resolution remain distinct problems as the industry approaches capabilities that earlier seemed theoretical.

The broader significance of Anthropic's approach lies in what it reveals about the evolving politics of AI legitimacy. As governments, religious institutions, and civil society organizations intensify their scrutiny of AI developers, companies are increasingly compelled to articulate not just what their systems can do but what values animate them. Anthropic's recruitment of theologians and ethicists, its engagement with faith communities, and its public refusals of morally contested contracts collectively constitute a strategy of normative credentialing — an attempt to earn trust through demonstrated commitment rather than purely technical achievement. Whether that strategy proves durable will depend on whether Claude's real-world behavior, and the company's business decisions, consistently reflect the principles it publicly espouses — a test that becomes more consequential with each advance in model capability.