The Vatican Picked Anthropic. Then Reality Hit.

Pope Leo XIV's Magnifica Humanitas encyclical and Chris Olah's appearance beside the pontiff on May 25 made headlines for their symbolism and positioning.

But what happened alongside it — and what it reveals — is a different story.

$500 Million. One Month. One Client.

Axios reported this week that an unnamed enterprise client ran up roughly $500 million in Claude charges in a single month after failing to put usage limits on employee AI licenses. Half a billion dollars. Gone.

The company has not been identified. But the mechanics of how it happened reveal the real problem.

In the old software world, a bad rollout meant paying for licenses that gathered dust. Static waste. Painful, but bounded. In the new agentic AI world — where models don't just answer questions but spawn tasks, retry failures, and operate autonomously — a bad rollout can become a financial catastrophe at speed.

Thousands of employees prompting, summarizing, refactoring, testing, spinning up agents. Each token costs money. Without usage caps, there is no natural ceiling. None were set.

Amazon's Tokenmaxxing Problem Makes It Worse

At almost the exact moment the Axios report dropped, Amazon was quietly shutting down an internal AI-usage leaderboard, according to ZeroHedge's reporting.

Why? Employees had started "tokenmaxxing" — deliberately routing unnecessary work through AI tools to inflate their scores on the leaderboard. Nobody was optimizing for productivity. They were optimizing for the metric.

This mirrors a basic principle of organizational behavior: when you measure a behavior, people perform it. The work becomes secondary.

Amazon is not publicly confirmed as the mystery $500M Claude client. But the parallel is striking.

What Mainstream Coverage Is Missing

Wired ran two pieces on the Vatican-Anthropic relationship. Forbes called the encyclical event "a positioning event for the AI industry" and noted Anthropic "won the dais." Religion News Service covered the ethics community's reaction.

None connected the Vatican story to the $500M bill story.

Here's what Leo XIV actually said in Magnifica Humanitas: AI creates conditions for a new form of slavery, where efficiency and surveillance serve the privileged few while everyone else suffers. He called for government regulation of private AI developers, worker protections, and retraining programs.

Then, the same week, an unnamed corporation proved his point by accidentally spending half a billion dollars in 30 days because nobody was managing it.

The Enterprise AI Hangover Is Real

Broader industry reporting notes that Microsoft has reportedly started canceling most Claude Code licenses and steering developers toward GitHub Copilot CLI instead. Uber, according to COO Andrew Macdonald, burned through its entire 2026 AI coding-tools budget by April — and Macdonald admitted it was "very hard to draw a line" between genuine productivity and AI-theater activity.

Uber blew a full year's AI budget in four months. Microsoft is pulling back on Claude licenses. An anonymous enterprise client just handed Anthropic $500 million by accident.

This is not a success story. It's a live experiment in what happens when corporations treat AI adoption as a performance metric rather than a tool.

Anthropic's Uncomfortable Position

Chris Olah said something striking at the Vatican, according to Wired's reporting. He told the audience: "Every frontier AI lab — including Anthropic — operates inside a set of incentives and constraints that can sometimes conflict with doing the right thing."

An AI company co-founder standing next to the Pope and admitting his own industry might be on the wrong path. That took guts.

But Anthropic is also the company whose usage-based pricing model just enabled someone to rack up $500M in a month. The company's Constitutional AI principles and its Vatican credibility don't automatically translate into guardrails on enterprise deployments. Those are two separate products.

Forbes contributor Gabriel Zainescu noted that the Vatican didn't invite OpenAI, Google DeepMind, or Meta to the encyclical presentation. It specifically chose Anthropic — and specifically Olah, the interpretability researcher. That's a deliberate bet on a particular lane of AI development.

The Vatican is making a calculated wager. It remains to be seen whether Anthropic can honor it.

What This Means for Regular People

If your company has rolled out AI tools on usage-based pricing with no spending controls, ask your CFO if there are caps. If nobody knows the answer, nobody is watching.

The enterprise AI bill that surfaced this week is a preview. As agentic AI — models that take autonomous actions, not just answer questions — becomes standard, the potential for runaway costs scales exponentially. This isn't a Claude problem specifically. It's a structural problem with how corporate America is deploying AI: fast, broad, and with no financial governance.

The Pope warned about AI creating systems where humans lose control. He was talking about humanity. He might as well have been talking about your company's IT budget.