The Tower of Babel Had No Audit Log

The Question Leo XIV Is Actually Asking

The encyclical opens with a stark binary: humanity faces "a pivotal choice: either to construct a new Tower of Babel or to build the city in which God and humanity dwell together."

Set aside the theology for a moment and look at the architecture of that metaphor. The Tower of Babel wasn't built by bad people. It was built by capable people with unified resources, unified technology, and unified direction — pursuing power without accountability, efficiency without humility, capability without limits. The Pope's warning about AI is structurally identical to that story: the danger isn't the technology, it's the absence of meaningful checks on who controls it and how.

He writes: "Never has humanity had such power over itself." And then asks the harder question — who actually holds that power, and what constrains them?

For AI agents, that question becomes: who authorized that action, and can you prove it?

The Babel Problem in Agentic AI

The encyclical's Babel image maps almost exactly onto the current trajectory of agentic AI: a single language (LLMs with converging capabilities), a single direction (autonomous execution at scale), and a design "conceived without reference to" external constraint. AI agents that can chain tool calls, write code, send messages, and modify infrastructure — without any meaningful human checkpoint — are building their own tower.

The Pope identifies the core pathology: power that is private and transnational, with "resources and the capacity to intervene that surpass those of many Governments." He's describing the large AI labs. But he's also describing what happens when an enterprise deploys an unconstrained AI agent with production credentials.

The problem isn't capability. It's the absence of walls.

What Meaningful Human Oversight Actually Looks Like

The encyclical calls specifically for "responsibility, transparency and the governance of AI." These words get used a lot. They're worth defining precisely in the context of agentic systems.

Responsibility means there is a named human being accountable for a consequential AI action — not a policy document, not a log entry after the fact, but a verifiable authorization that happened before the action executed. In practice, this means hardware. A physical authenticator that a human being touched, producing a cryptographic signature bound to the specific intent of that action.

Transparency means the decision logic is legible. What can an agent do autonomously? What requires logging? What requires a physical human gesture? These thresholds should be explicit, auditable, and adjustable — not buried in model weights or vendor terms of service.

Governance means those thresholds are actually enforced. Not aspirationally. Not by convention. By a system that cannot be bypassed without a hardware-rooted human act.

This is the gap in most current deployments. Enterprises are running AI agents with broad permissions and soft guardrails — rate limits, content filters, policy prompts. Those are speed bumps, not walls.

Three Things the Pope Says Must Be Protected

Chapter Four of Magnifica Humanitas frames the stakes around three values: truth, work, and freedom. Each one maps directly to the challenge of agentic AI authorization.

Truth: AI agents that act without verifiable authorization produce outcomes with no legible chain of accountability. Who approved the deployment? Who authorized the data access? When did a human last review what the agent was doing? Hardware-rooted authorization creates an audit trail that is cryptographically non-repudiable — not a log you can edit, but a signed record tied to physical presence.

Freedom: The encyclical warns against AI systems that create "dependencies and societal control" — systems that subtly capture human decision-making until autonomy feels impossible. The antidote is graduated oversight: routine, low-stakes actions can be automatic; consequential, irreversible actions require a human in the loop. The key is that the degree of human involvement is transparent and tunable, not hidden inside a black box.

Work: The Pope defends meaningful human participation in consequential processes as having intrinsic value — not as nostalgia, but as a structural claim about dignity. Requiring a human to physically authorize a high-stakes AI action isn't just a security control. It's an assertion that consequential decisions require human presence. That someone, somewhere, is actually responsible.

The Nehemiah Parallel

The encyclical's second biblical image — Nehemiah rebuilding Jerusalem's walls after the exile — is the one that resonates most. The people returning to a ruined city didn't abandon it. They assessed the damage and rebuilt with purpose. Nehemiah surveyed the ruins at night before proposing a plan.

We're in a similar moment with agentic AI. The city is already being built — fast, at scale, by actors with enormous resources. The question isn't whether to build. It's whether the walls go up before or after the next major incident.

The walls are: hardware-rooted authorization. Intent-bound signatures. Audit trails that cannot be edited after the fact. Graduated human oversight with real enforcement teeth.

Bottom Line

The Pope's encyclical asks: who holds power over AI, and what actually constrains them?

Soft controls — policies, prompts, rate limits — are not the answer. The answer is hardware. Physical human presence, cryptographically verified, bound to specific intent, logged and auditable.

That's not a theological position. It's an engineering requirement. And it's one the industry is only beginning to take seriously.