The Vatican just published 245 paragraphs arguing that your training data isn't yours to sell, and brought a man currently suing the Pentagon to help make the case.

The Summary

The Signal

The Vatican doesn't do hot takes. When it publishes 245 paragraphs on artificial intelligence, it's not reacting to the news cycle. It's attempting to set the moral framework for the next century. Pope Leo XIV's first encyclical on AI does something Silicon Valley hates: it argues that the infrastructure of intelligence has moral weight, and that weight belongs to humanity, not whoever scraped the data first.

The core claim cuts deep. By calling data a "common good", the encyclical rejects the foundational premise of the agent economy as currently constructed. If training data is common property, then the companies sitting on exabytes of human output don't own the raw material of intelligence. They're tenants, not landlords. This isn't a regulatory framework. It's a philosophical challenge to the property rights underlying every foundation model.

"The Vatican is arguing that the companies building AI don't own the substrate of thought itself."

The choice of Christopher Olah as Vatican lay speaker isn't symbolic window dressing. Olah co-founded Anthropic, which split from OpenAI over safety concerns and is now actively suing the Trump administration over military AI applications. You don't pick that guy if you want a safe, consensus take. You pick him if you want to signal that AI safety and AI ethics are inseparable from AI governance, and that the people building the tools have a responsibility beyond shareholder returns.

The timing matters. We're in the window where AI governance is still being negotiated. The EU has its AI Act. The US has executive orders and vibes. China has its own model. But no one has a coherent moral framework that spans cultures and legal systems. The Catholic Church has 1.4 billion members and 2,000 years of practice building institutions that outlast empires. When it publishes an encyclical, it's not trying to win this quarter. It's trying to set the terms for the argument that shapes the next 50 years.

Key implications for the agent economy:

  • If data is a common good, agent training becomes a use case, not a property claim
  • The moral neutrality argument (we just build tools, users choose how to use them) gets rejected at the institutional level
  • Companies that align with this framework get moral and social capital; companies that don't get painted as extractive

The encyclical doesn't ban AI or call for a pause. It does something harder: it argues that technology is never morally neutral, which means every architectural choice is an ethical choice. Centralized vs. decentralized training. Proprietary vs. open weights. Opt-in vs. opt-out data collection. These aren't engineering tradeoffs. They're moral decisions with consequences for billions of people who never consented to being in the training set.

The Implication

Watch how this plays in jurisdictions where the Catholic Church still moves policy. Latin America, parts of Europe, the Philippines. If "data as common good" becomes the default framing, it doesn't just affect foundation model companies. It affects every agent that scrapes, every platform that monetizes user-generated content, every tokenization play that treats digital exhaust as private property.

For builders in the agent economy, the strategic move is to design systems that assume shared data ownership from the start. Consent-based training. Revenue sharing when agents use your data. Provenance tracking that respects contribution. These aren't nice-to-haves. They're the price of entry to markets where the moral framework just shifted under your feet.

Sources

BeInCrypto | RWA Times | Decrypt