Pope Leo XIV Calls to ‘Disarm’ AI — The Vatican Just Entered the AI Governance Ring
Pope Leo XIV published his first encyclical, Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence, and it’s not a gentle suggestion — it’s a radical intervention. The Pope called for AI to be “disarmed”, warned that AI concentration in a handful of US tech firms threatens human dignity, and cited Gandalf in his argument for epistemic humility.
Anthropic’s cofounder joined Leo on stage at the Vatican for the launch. Chris Olah was there too, saying AI “must be guided from outside Big Tech.” The encyclical targets the concentration of compute and power — not just AI’s risks, but who controls it.
The Vatican has diplomatic reach, moral authority with 1.4 billion Catholics, and an institutional history of weighing in on issues of human dignity (Rerum Novarum, Laudato Si’). This isn’t a fringe opinion — it’s a new, powerful voice in the AI governance conversation.
Why it matters: The AI governance debate has been dominated by the US, EU, and China. The Vatican just introduced a fourth pole — one with moral suasion that neither trade blocs nor tech CEOs can match. If AI treaties ever materialise, expect the Holy See at the table.
Anthropic Accidentally Leaked Details of Its Next Model
In an embarrassing security lapse, Anthropic left roughly 3,000 unpublished assets publicly accessible via its CMS — including details of an unreleased AI model, an invite-only CEO retreat, images, PDFs, and internal data. Cambridge cybersecurity researcher Alexandre Pauwels assessed the trove at Fortune’s request after discovering the data was accessible without a login.
Anthropic attributed the issue to “human error in CMS configuration.” The company secured the data after Fortune notified them. The CMS architecture apparently made all assets public by default unless explicitly set private — a design choice that seems remarkably lax for a frontier AI lab.
The unreleased model is believed to be a significant iteration — potentially what Anthropic has referred to as a “step change” in capabilities. The leak also revealed details of an exclusive CEO event, suggesting Anthropic is courting top leadership while simultaneously struggling with basic security hygiene.
Why it matters: A company whose entire pitch is “safety-first AI” left thousands of documents open by default. The mismatch between Anthropic’s safety rhetoric and its operational security is becoming a pattern — from Mythos emerging as a hacking tool, to supply chain fights with the Pentagon, to now this.
AI Warfare Is Already Here — Even Anthropic Knows Its Red Lines Won’t Hold
The Verge published the definitive account of the Anthropic-Pentagon standoff — and it’s bleaker than the headlines suggest. Reporter Hayden Field traces AI’s military trajectory from Project Maven (2017) through to today, where Anthropic is the only military AI contractor drawing meaningful limits. But here’s the kicker: even Anthropic reportedly believes its two red lines — no domestic mass surveillance, no fully autonomous lethal weapons — won’t hold long-term.
The piece reveals that Defense Secretary Pete Hegseth demanded contract renegotiation in January 2026, eliminating previously agreed grey areas. The Pentagon then designated Anthropic a military supply chain risk in March, and Trump banned all government agencies from using Claude. The relationship has warmed slightly with Anthropic’s Mythos cybersecurity model, but the fundamental tension remains.
UC Berkeley’s Andrew Reddie put it plainly: “We’ve kind of crossed the Rubicon while we pretend that we haven’t.”
AlphaProof Nexus: AI Solves Decades-Old Math for a Few Hundred Dollars
Google DeepMind’s AlphaProof Nexus autonomously solved 9 out of 353 open Erdős problems, proved 44 of 492 OEIS conjectures, and settled a 15-year-old Hilbert function question. The system uses an LLM + Lean compiler feedback loop — and the surprise finding was that the simplest agent (LLM + compiler feedback alone) could also solve all 9 Erdős problems. Inference cost: a few hundred dollars per problem. This is the “shift from specialised trained systems toward simple agentic loops” that keeps showing up in 2026.
Uber Exhausted Its 2026 AI Budget in 4 Months
Uber president Andrew Macdonald went public: the company can’t link rising AI token consumption to consumer value. After rolling Claude Code out to 5,000 engineers in January, the entire annual AI budget was gone by April. “That link is not there yet,” Macdonald said. CEO Dara Khosrowshahi’s response: hire fewer humans. We wrote about this in depth — Uber Burned Through Its AI Budget in 4 Months.
Intuit Cuts 3,000 Jobs — But Says It’s “Not About AI”
Intuit laid off 17% of its workforce — about 3,000 people — and CEO Sasan Goodarzi insisted the cuts had “nothing to do with AI.” The restructuring is framed as simplifying operations. But when a company that just rolled out enterprise AI tools cuts 17% of staff in the same breath, the market reads between the lines.
Andrej Karpathy Joins Anthropic
OpenAI co-founder and former Tesla AI director Andrej Karpathy has joined Anthropic on the pre-training team. It’s a significant talent signal — Karpathy is one of the most visible AI researchers in the world, and his move from OpenAI → Tesla → Eureka Labs → Anthropic tracks the industry’s consolidation toward a few major players.
Judge Blocks Pentagon’s Anthropic Supply Chain Designation
A federal judge has temporarily blocked the Pentagon’s designation of Anthropic as a military supply chain risk — the same designation the Trump administration used to ban government agencies from using Claude. The ruling is a temporary reprieve in Anthropic’s ongoing legal battle with the DOD.
🔍 THE BOTTOM LINE
Three threads this week that should worry anyone paying attention: AI warfare red lines are eroding faster than anyone admits, enterprise AI spending can’t justify its own existence yet, and the people building frontier models are publicly saying they can’t guarantee safety. The through-line? Nobody’s in control — and the people who should be are telling us so themselves.