Vatican Releases AI Governance Framework With Input from Anthropic Co-Founder

The Vatican has released an 82-page encyclical on artificial intelligence, establishing a formal framework for managing the technology's operational and ethical challenges. The document, titled "Magnifica Humanitas" and signed by Pope Leo XIV on May 25, 2026, provides actionable criteria for assessing AI's impact on society, labor, and geopolitics. In a notable move, Anthropic co-founder Chris Olah joined the Vatican’s official presentation, signaling a serious dialogue between a major AI lab and a global institution focused on ethical governance. The encyclical aims to move the conversation beyond abstract principles and toward concrete risk management for companies and developers.

The document argues that technology is never neutral, but instead reflects the values of its creators. It warns against a "technocratic paradigm" where efficiency and profit are prioritized over human dignity. To make this actionable, the encyclical offers a model for evaluating AI projects: they can either become a "Tower of Babel," driven by unchecked ambition that erodes trust, or a "rebuilding of Jerusalem," a collaborative effort that reinforces social bonds. At its core, the framework is built on principles like human dignity, solidarity, and shared responsibility, positioning them as essential standards for any ethical AI system.

Frontier Models Need 'Moral Voices,' Says Anthropic Co-Founder

Speaking at the launch, Anthropic's Chris Olah essentially endorsed the encyclical's call for independent oversight. He argued that all frontier AI labs are driven by incentives like commercial pressure, geopolitical competition, and simple ambition, all of which can conflict with safety. According to Olah, this creates a critical need for "informed critics" and "moral voices that the incentives cannot bend" to provide guardrails that internal teams are often too biased to build themselves.

Olah highlighted three key areas that require external ethical judgment, but his most striking comments were about the nature of frontier models themselves. He said that research into their internal structures has revealed phenomena that are "mysterious, even unsettling." Olah noted his team has found internal processes that mirror findings from human neuroscience, evidence of introspection, and states that function like human emotions, including joy, fear, and grief. "I don’t know what that means, but I think it warrants ongoing discernment," he admitted. For businesses, this raises a huge red flag: deploying systems with unpredictable and unexplainable internal states creates massive operational risk, making third-party audits and red-teaming an essential compliance step. Olah also pointed to the need for external guidance on managing large-scale labor displacement and ensuring AI is integrated into society in a way that supports human flourishing, not just corporate goals.

The encyclical tackles this problem of unexplainable models head-on. It cautions against a narrow focus on "alignment" without first establishing a social consensus on what values AI should be aligned with. The document insists that responsibility must be clearly traceable throughout an AI system's entire lifecycle. Opaque, black-box systems make it impossible to assign liability or offer remedies when things go wrong, which necessitates new legal and technical frameworks that mandate transparency and human control. The Vatican also extended its critique to the AI supply chain, from the mining of minerals to the often-exploitative human labor used for data labeling, calling this "invisible" work a "new form of slavery" and demanding greater corporate due diligence.