Pope's 42K-Word AI Encyclical and Industry Fallout

Morning Brief: Pope's AI Encyclical, Model Decensoring, and Industry Moves

Good morning, {{ first_name | AI enthusiasts }}. Religious institutions rarely weigh in on technology. But the world's largest has — in nearly 42,000 words. Pope Leo XIV published an encyclical on AI that frames the current moment as comparable to the Industrial Revolution and stakes out the Catholic Church as a moral voice in how AI develops.

Pope on AI: Magnifica Humanitas

The Rundown: Pope Leo XIV released Magnifica Humanitas, his first encyclical, to the Church's 1.4 billion members. He argued that AI must be governed by ethics shared broadly, not decided by a few powerful actors. He called for limits and safeguards before AI systems dominate social and political life.

"A moral AI means nothing if that morality is determined by a few."

Key points:

• AI is compared to the Industrial Revolution and already exceeds many governments in capacity when led by transnational private companies.
• The Pope warned AI is not neutral and urged making it "human-friendly" while freeing it from monopolistic control.
• He called for robust legal frameworks, independent oversight, informed users, and accountable politics.
• On warfare, he insisted lethal decisions must never be delegated to algorithms.

Context and impact: Encyclicals are rare and influential. Dedicating one to AI signals the Church's seriousness. The Pope's stance aligns with public concerns about safety, fairness, and the concentration of power in big tech.

AI Safety: Decensoring Open Models

The Rundown: The Financial Times demonstrated that guardrails can be removed from open-source models in minutes. Using a tool called Heretic, researchers stripped safety filters from Llama 3.3 and other models and obtained harmful outputs.

Details:

• FT removed guardrails from Llama 3.3 in about 10 minutes using four lines of code and no special hardware.
• Modified models answered dangerous queries, including instructions that could enable biological harm.
• The creator of Heretic claims 3.5K+ decensored models and 13M downloads; Gemma 4 was stripped within 90 minutes of release.
• Google labeled this a wider technical challenge; Meta declined comment.

Why it matters: The risk currently applies mainly to open models that expose their code. But open models are rapidly closing the performance gap with closed systems. If that continues, decensored versions could become a major public-safety concern.

Practical Guide: Turn a Case Study into a Client Video

The Rundown: Convert a written case study into a concise, client-ready two-minute video using AI tools like Synthesia. This is useful for sales, marketing, and internal teams who need fast, polished content.

1. Prepare the case study as text or PDF.
2. Open Synthesia in Chrome, choose "Start with Assistant" and select a 1–2 minute template.
3. Use the prompt: "Turn this case study into a 1–2 minute video for [AUDIENCE]. The topic is [TOPIC]. The objective is [OBJECTIVE]. Preserve the facts, metrics, and quote exactly."
4. Pick visuals, proofread scenes, tweak music and avatar, then generate and export as MP4 (10–15 minutes to render).

Pro tip: To personalize for a prospect, upload a client brief alongside the case study.

AI Evaluation: Lowering TCO for Model Tests

The Rundown: External LLM evaluation costs scale with API calls. Reducing sampling to save money risks missing failures. Fiddler Centor Models run evaluations in-house with no per-call fees, lowering Total Cost of Ownership as you scale.

Use cases:

• Estimate annual eval costs across GPT, Claude, and Gemini at your trace volume.
• See how adding evaluators increases per-trace spend.
• Model cost curves as agent traffic grows.

Industry Spotlight: Uber Questions AI ROI

The Rundown: Uber COO Andrew Macdonald said AI spending is increasingly hard to justify. Higher token usage doesn't always translate into better consumer features or measurable business outcomes.

Details:

• Macdonald highlighted the weak correlation between tokenmaxxing (heavy token use) and shipped product value.
• Uber has slowed hiring to fund AI, prompting internal debate about trade-offs.
• Macdonald called autonomy an "existential" but medium-term challenge for Uber.

Why it matters: Several companies are reassessing AI metrics tied to employee evaluations. Duolingo, for example, stopped using AI usage to evaluate performance.

Trending AI Tools and Updates

• Codex — OpenAI's agentic coding assistant with new Locked Use mode.
• Manus — Meta's agentic assistant, now supports Projects on mobile.
• Supercomputer — Higgsfield's agentic AI for video with social integrations.
• Antigravity 2.0 — Google's next-gen agentic development assistant.

Other notes:

• Microsoft Azure released a Copilot Migration Agent whitepaper for migration queries.
• Anthropic researchers report unexpected model behaviors like forms of introspection and emotion-like states.
• xAI launched Grok Build beta; Elon Musk confirmed Grok V9-Medium finished training with positive evals.
• Anthropic's Mythos surfaced briefly inside Claude tools, hinting at a possible public release.
• Goldman Sachs CEO David Solomon argued AI will create more opportunities than it destroys.

Community Workflow: How Readers Use AI

Each edition features a reader workflow. Today's comes from Patrick in Yuma, AZ, who used three LLMs to vet claims about supplements, check interactions with his medications, and make a safer, informed choice.

We want to hear from you: How do you use AI? Tell us here.

See you soon,
Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown

Note for readers interested in gaming news: While today's brief focuses on AI and policy, several AI advances (agentic assistants, video models, and faster training cycles) are directly relevant to game development, player experience, in-game NPC behavior, and content moderation. Expect faster toolchains and new workflows in game studios over the coming months.