On May 25, 2026, Pope Leo XIV stood in Vatican Synod Hall alongside Anthropic co-founder Chris Olah, before a room packed with cardinals, theologiansand AI researchers, and presented Magnifica Humanitas, the Catholic Church’s first formal encyclical on artificial intelligence.
The Catholic Church weighing in on AI might read like the setup of a poorly-constructed joke, but it isn’t. Technology is never neutral. It bears the fingerprints of whoever designs, funds and deploys it. Pope Leo XIV just put that on the record, and the Vatican called it before Silicon Valley will. And maybe that’s what it will finally take to start an earnest, widespread movement for AI regulation and guidelines.
What the Encyclical Actually Says
Strip away the scripture and the manifesto reads like a working draft from the EU AI Act, NIST or ISO 42001, addressing power concentration, transparency, labor displacement, synthetic disinformation and human dignity. The Pope is giving a moral backbone that legislators, companies and leaders can lean on for AI governance enforcement.
The encyclical acknowledges that a handful of private, transnational actors now possess the capacity that surpasses that of many governments. Olah acknowledged it from the stage, noting that AI development is concentrated in a handful of wealthy nations with no mechanism to share the gains. The manifesto then calls for independent checks, real visibility into the algorithms, fair access to data and a path to recourse when things go wrong. An entire section addresses the dignity of work through the digital transition.
All of this is framed by the debate of Babel versus Nehemiah. Should humanity build society in the manner of Babel, where humans built a tower, optimized for scale that was answerable to no one, or a wall in the style of Nehemiah, with every section owned by someone and each decision accountable to the people it affects?
AI is existential. We already know this, but the Pope is insistent that humanity takes it seriously, especially those in power who are actively making these decisions that impact everyone. It matters because even the Pope isn’t trying to stop innovation or the advancement of technology. He is calling for everyone to be considered while we adjust society to the powerful technology, which others have been calling for for years, without being taken seriously, or at least not heard when Big Tech money, influence and lobbying win.
Where the Industry Is Falling Short
The reality is, we need AI governance that we agree on as a society. Somewhere right now, someone is presenting a slide titled “Responsible AIFramework,” while the model they are using is encoding discrimination. The tech industry has gotten very good at calling that governance, but it’s not.
The shortfall is structural because ethical consideration typically lives in the legal department or the post-incident review and never reaches the model before deployment. And accountability keeps getting downgraded, like what happened to New York’s RAISE Act, where the initially proposed mandatory audits became optional and independent oversight became self-certification by the time the lobbyists were done. That pattern has repeated across every major AI governance effort.
At the Vatican, Olah admitted every frontier AI lab operates inside incentives that can conflict with doing the right thing, then asked the Church and civil society to serve as the watchdog, reinforcing that influence and enforcement are not the same. The gap between the framework and the system is where harm lives.
Why Governance Is the Competitive Advantage
The framing of governance as a balance against innovation assumes the two pull in opposite directions, and that is the most creatively expensive myth in the industry. The companies that embed accountability into development ship faster in the long run because they are not rebuilding under a consent decree. Procurement teams at enterprise buyers can actually say yes to these companies. The leaders who do not want their name on the next scandal actively seek them out.
The first movers on AI governance standards like ISO 42001 are not slowing down. They are building the trust moat that their competitors will spend the next three years and a few lawsuits trying to cross. The encyclical’s Nehemiah model is good operating discipline dressed in scripture: shared responsibility with local accountability. However, the Pope won’t get companies to change alone. Leaders have to make an active choice to prioritize governance for the benefit of all.
What Technology Leaders Can Do Now
None of what follows requires a transformation program or additional headcount. It requires deciding that the model in production matters more than the framework on the wall.
An honest absence audit is the right place to begin. For the most consequential AI system a company is using, ask who was missing from the training data, the design room and the test population, and write down the answer even when it stings. From there, move governance out of the legal department, because an accountability check belongs inside the development workflow before launch, not after.
Power concentration in the stack deserves a hard look. Leaders need to understand who owns the models and infrastructure a company depends on, and ask: what are the actual terms, and what is the plan when those terms change? Leaders must demand transparency that explains the model logic, dataand demographics against which a system was tested. Treat “proprietary” as the start of a negotiation.
Most organizations underinvest in the monitoring infrastructure to know what a model is doing right now, not what it did at certification. A model is not a building that gets inspected once and signed off on. It drifts, degrades and surprises organizations in production, and the harm surfaces exactly where an annual audit cannot see it.
Governance done once a year is a photograph of a car that has already moved. Behavioral baselines established at deployment, outcome monitoring at the population level rather than just the aggregate and escalation paths that reach someone with the authority to act are what catch failures before they cause real damage.
The point isn’t dashboards for dashboard’s sake. Instead, companies need to build for security and adversarial behavior, because a model that can be manipulated is a model that isn’t actually under control. The organizations building observability and governance muscles now are the ones who will never have to explain, after the fact, why they did not see a problem coming.
Where That Leaves Us
The Pope will not write a single line of statute. What an encyclical does is move the Overton window, making AI governance policies more publicly acceptable. It hands legislators, regulators and boards moral cover to say the thing lobbyists have spent years making politically expensive to say: that self-certification is not accountability, and that concentrated, unaccountable AI power is a problem worth governing. Nine in ten U.S. policy insidersalready say AI must be regulated. When the diagnosis ethics practitioners have been making for years gets restated from that pulpit, it stops sounding like activism and starts sounding like consensus, and that changes what a regulator can credibly propose.
Moral authority is not enforcement. Words from the Vatican will not hollow out lobbying budgets, and only people in the arena will change that. But if 40,000 words can finally make the boardroom hear what practitioners have been saying for years, that is an assist worth taking.
So, will we as society ultimately choose to build in the way of Babel or Nehemiah? Leaders are already picking one in every procurement decision, model audit and board vote and the ones who have not named their choice yet will have it named for them when the regulation arrives.
