The Information Machine

2026-05-27

Pope Leo XIV's 'Magnifica Humanitas' frames AI as civilization's defining binary while The Verge reports evidence of AI-generated text within the encyclical itself, and YouTube deploys automated AI video detection at platform scale.

What

Pope Leo XIV published 'Magnifica Humanitas' [1], the Catholic Church's first encyclical dedicated to AI, framing the technology as a choice between 'Babel' (power concentration and enforced uniformity) and 'Jerusalem' (technology co-built for the common good), with explicit calls for explainable AI in consequential decisions and regulation preventing power concentration. The Verge separately reported evidence that AI-generated text may appear within the encyclical itself [2], a potentially self-undermining detail for a document whose central concern is AI's effect on human authenticity and authority. A secondary controversy surrounds the document's origins: Anthropic co-founder Christopher Olah's advisory role in shaping the text drew accusations that a tech company effectively laundered its own product framing through Vatican doctrine [3]. YouTube announced it will use 'new internal signals' to automatically flag photorealistic AI video content [4], adding the world's largest video platform to the AI content provenance debate at scale and raising the question of whether that deployment integrates the C2PA coalition standard or operates as a parallel proprietary track. Anthropic's annualized revenue reportedly grew from $14B at its February funding round to approximately $30B or more by April [5], with some sources citing $44B and a doubling rate of every six weeks, as both Anthropic and OpenAI continue to expand enterprise deployment infrastructure.

Why it matters

The papal encyclical story is simultaneously a policy document advocating AI regulation, a case study in how tech companies influence institutional AI governance narratives, and — if the AI-generated-text reporting holds — a live demonstration of the authenticity concerns the document itself raises, making it an unusually self-referential moment in the AI governance debate. YouTube's platform-scale AI video detection matters because it will set the practical norm for how billions of users encounter AI content labeling, regardless of whether its technical architecture connects to the cross-industry C2PA coalition or diverges from it.

Open questions

  • If The Verge's evidence that AI-generated text appears within 'Magnifica Humanitas' [2] is confirmed, what does that mean for the document's authority as an AI governance guide — and for Anthropic's implicit endorsement through Christopher Olah's shaping role [3]?

  • YouTube's 'new internal signals' for AI video detection [4] leave open whether the world's largest video platform is integrating the C2PA/SynthID cross-industry coalition architecture or deploying a parallel proprietary track — and whether fragmentation between approaches will undermine the coalition's goal of universal provenance standards.

  • With Anthropic's annualized revenue reportedly doubling from $14B to $30B or more in roughly two months [5], does the pace of commercial growth change the sustainability calculus around its $1.25B/month compute obligations, or does the wide variance across reported figures make them unreliable as a planning anchor?

  • ICE and CME Group have both announced plans to launch GPU futures markets while analysts describe AI tokens as potentially 'the first true digital store of value' [6] — if compute becomes a traded commodity at the scale of energy markets, how does that reshape who controls AI infrastructure access and which actors can afford frontier model development?

Thread movements (4)

  • papal-ai-encyclical — Pope Leo XIV published 'Magnifica Humanitas' [1], the first papal encyclical on AI, framing development as Babel versus Jerusalem and calling for explainable AI and regulatory checks on power concentration; The Verge reported evidence of AI-generated text within the document itself [2], and Anthropic co-founder Christopher Olah's advisory role drew accusations of tech-company doctrine-laundering through the Vatican [3].
  • ai-content-provenance-watermarking — YouTube announced it will use 'new internal signals' to automatically flag photorealistic AI video content [4], adding the world's largest video platform to the content provenance space at scale while leaving open whether those signals integrate the C2PA open standard or SynthID specifically.
  • openai-codex-enterprise-rollout — New items add further enterprise momentum to OpenAI Codex's documented rollout [7][8], with Anthropic's competing revenue trajectory [5] providing independent context on how the enterprise AI coding market is distributing between the two leading platforms.
  • anthropic-agent-ai-direction — Reporting on Anthropic's revenue trajectory [5] — from $14B ARR in February 2026 to approximately $30B or more by April, with some sources citing $44B and a doubling rate of every six weeks — adds quantitative texture to the sustainability question around the company's $1.25B/month compute obligations, though the figures carry wide variance and lack independent verification.

Notable items (2)

  • 🟡 Machine earning
    Semafor Technology
    A Semafor Technology newsletter roundup [6] surfaces several structurally significant signals in a single dispatch: ICE and CME Group are both launching GPU futures markets with projections of a $6 trillion commodity market rivaling energy; Robinhood launched AI agents for autonomous retail stock trading; China is restricting overseas travel for top AI employees at Alibaba and DeepSeek similarly to nuclear scientists; and UC Berkeley Law banned AI use across all stages of legal assignments — each a notable institutional response to AI's expanding role that warrants separate tracking.
  • Full automation of AI R&D probably yields a large speed up even without a software-only singularity
    Alignment Forum
    Ryan Greenblatt's Alignment Forum analysis [9] argues that full automation of AI R&D produces a one-time speedup equivalent to roughly 2.5–3.5 years of progress in the first post-automation year even without a software-only singularity, because automated AI researchers simultaneously improve the AI labor force and scale the researcher count — a quantitative frame for takeoff timelines that challenges the assumption that subcritical recursive improvement coefficients imply modest acceleration.