Pope Seeks To Reshape Global AI Agenda With Historic Manifesto
May 25, 2026
D.A.D. today covers 10 stories from 3 sources. What's New, What's Innovative, What's Controversial, What's in the Lab, and What's in Academe.
The Daily AI Digest is a daily AI briefing automated by Alexander Panetta — a veteran political journalist tracking the field during a Master's in AI Management at Georgetown University.
D.A.D. Joke of the Day: My company replaced our IT guy with AI. Now when something breaks, it apologizes for 200 words before suggesting I turn it off and on again.
What's New
AI developments from the last 24 hours
Pope's 42,000-Word AI Manifesto—Drafted With Anthropic—Sets Global Ethical Agenda
Pope Leo XIV released Magnifica humanitas Monday, a sweeping papal encyclical on artificial intelligence co-presented with Anthropic co-founder Christopher Olah—a deliberate signal of dialogue between spiritual and tech leadership. The 42,300-word document calls for government regulation of AI companies, legal protections for workers displaced by automation, critical AI education in schools, and a firm requirement that humans—not algorithms—retain responsibility for all weapons decisions. The Pope argues AI must not concentrate power among a few or sacrifice jobs for profit, warning that material progress without human purpose produces social instability.
Why it matters: When the leader of 1.4 billion Catholics frames AI governance in terms of human dignity, worker protection, and accountability—and does so standing next to an Anthropic executive—it signals that the public and political pressure on AI companies is entering a new phase. Executives who've been navigating a mostly technical debate should expect the conversation to become explicitly moral.
Source: vaticannews.va · Source: nytimes.com
George Hotz Warns AI Coding Agents Will Become Enterprise's Costliest Mistake
George Hotz, the hacker known for first jailbreaking the iPhone and founding Comma.ai, argues that AI coding agents will become "one of the most costly mistakes in software history." After six months of personal testing across multiple models and approaches, Hotz claims agents produce code that mimics real programming but breaks in ways that grow harder to detect over time. His sharpest point: large organizations will suffer most because their slower feedback loops let AI-generated errors compound, while solo developers catch problems faster.
Why it matters: This is one developer's opinion based on personal experience—not research—but it articulates a concern quietly shared by some practitioners: that AI coding tools may create technical debt faster than they create value, especially at enterprise scale.
Discuss on Hacker News · Source: geohot.github.io
Memory Costs Now Drive Two-Thirds of AI Chip Budgets
Memory now accounts for nearly two-thirds of AI chip costs, up from about half just two years ago. Epoch AI's analysis found HBM (high-bandwidth memory) grew from 52% to 63% of component costs between Q1 2024 and Q4 2025, while logic processors held steady around 13%. Total industry spending on AI chip components more than doubled to $52 billion in 2025, with memory alone driving $20 billion of that increase.
Why it matters: This cost structure explains why running AI locally remains expensive—and why memory manufacturers like SK Hynix and Samsung hold unusual leverage over the AI hardware supply chain.
Discuss on Hacker News · Source: epoch.ai
Australian Four-Day Workweek Trial: 15 Companies, Zero Productivity Drops
A peer-reviewed study in Nature's Humanities and Social Sciences Communications tracked 15 Australian companies that adopted a four-day work week (full pay, 80% hours, same output expectations) between 2022 and 2024. Results: 14 of 15 continued the model after the trial. Six reported productivity increases, nine said it stayed flat, and none reported declines. Companies measured success through revenue, profit, project completion rates, staff turnover, and absenteeism.
Why it matters: This adds longitudinal evidence to the four-day week debate; the retention rate (93% continuing) and zero productivity losses may give executives considering the model more confidence in the business case.
Discuss on Hacker News · Source: scienceaim.com
Why AI Assistants Make Poor System Architects
Software consultant Charlie Holland argues that organizations are misusing AI assistants as system architects—a pattern he says he's observed at three companies recently. His critique: AI tools are "pathologically agreeable," offering plausible-sounding designs without the critical pushback that experienced architects provide. They pattern-match against training data rather than understanding team constraints, organizational politics, or when to reject unnecessary complexity. The piece is opinion based on anecdotal observation, not research.
Why it matters: As AI coding tools expand from writing code to designing systems, the question of where to draw the line on AI authority is becoming a live debate in engineering leadership.
Discuss on Hacker News · Source: hollandtech.net
DeepSeek Makes 90% Price Cut Permanent, Undercutting OpenAI Long-Term
DeepSeek announced its 75% discount on the V4 Pro API will become the permanent price after May 2026. Final pricing: $0.435 per million input tokens (cache miss), $0.87 per million output tokens—roughly 90% cheaper than GPT-4o's standard API rates. Both V4 Pro and V4 Flash support 1 million token context windows. Early users report strong value; one noted $1.50 covered 65 million tokens for coding tasks.
Why it matters: DeepSeek's aggressive pricing could force Western labs to respond, giving teams another low-cost option for high-volume API work and potentially reshaping the economics of AI-powered products.
Discuss on Hacker News · Source: api-docs.deepseek.com
What's in Academe
New papers on AI and its effects from researchers
Loan Officers Deliberately Avoid AI Explanations When Bonuses Are at Stake
A new NBER study found that people acting as loan officers actively avoided looking at AI explanations when their bonuses depended on loan repayment—even while still using the AI's predictions. The research, using real $10,000 loans, showed participants would override profit-maximizing AI recommendations when explanations revealed the model penalized non-White or female borrowers. But when money was on the line, they sought the predictions while deliberately skipping the explanations. The avoidance behavior diminished when explanations were framed as purely financial or when demographic information was hidden.
Why it matters: This is evidence that "explainable AI" requirements may backfire—users who face financial pressure will game the system by taking AI outputs while strategically ignoring the explanations that would reveal uncomfortable truths.
Deep Learning Model Claims 10x Better Crisis Prediction for Financial Regulators
An NBER paper finds that graph-based deep learning applied to $40 trillion in security-level portfolio holdings data substantially outperforms traditional methods for predicting how financial institutions will trade during stress events. The model claims more than ten times the explanatory power for asset return variations during crises compared to existing approaches. Notably, the architecture works even when entire asset classes or investor types are excluded from training—suggesting regulators could assess risks in novel market conditions.
Why it matters: This is academic research, not a deployed tool, but it signals that AI-driven systemic risk monitoring could give regulators earlier warning of financial contagion—relevant context as policymakers debate AI's role in oversight.
Time-Locked AI Models Could Finally Fix Finance's Backtesting Problem
Researchers trained language models exclusively on text available up to specific calendar dates—no future information allowed—and found that scaling these 'point-in-time' models to 4 billion parameters nearly closes the performance gap with conventional models trained on all available data. Portfolios built using these chronologically honest models achieved strong risk-adjusted returns, demonstrating they can extract meaningful signals without the lookahead bias that plagues backtesting. Models were benchmarked monthly from 2013-2024.
Why it matters: For quantitative finance teams, this suggests AI-driven investment research can be backtested with genuine historical fidelity—a persistent challenge when standard models inadvertently 'know' future information.
Persuasive AI Explanations May Hurt Decision-Making, Study Finds
A large-scale behavioral study found that persuasive AI explanations don't actually improve human decision-making—and may make it worse. Researchers tested how narrative explanations from LLMs affected people performing classification tasks. Elaborate, persuasive explanations performed no better than simple AI predictions alone. Worse, the narratives increased reliance on AI for both correct and incorrect predictions. Exploratory analysis suggested more persuasive explanations may have slowed response times and made it harder for people to distinguish good AI predictions from bad ones.
Why it matters: For teams using AI assistants that explain their reasoning, this challenges the assumption that more elaborate justifications help—they may actually increase blind trust in flawed outputs.
Framework Aims to Standardize What 'Explainable AI' Actually Means
A new preprint proposes a framework for making AI explanations more systematic and measurable. The researchers argue that high-level goals like fairness, accountability, and trust aren't independent—they form dependency chains where some properties must exist before others can be achieved. Their three-axis taxonomy (who needs the explanation, what it's for, how it's justified) aims to translate abstract concepts into concrete, testable benchmarks. The work is conceptual, based on literature review rather than empirical testing.
Why it matters: As AI regulation tightens globally, companies need standardized ways to demonstrate their systems are explainable—this framework could help turn vague compliance requirements into checkable criteria.
What's On The Pod
Some new podcast episodes
The Cognitive Revolution — All Compute Is Food: Palisade's Jeffrey Ladish on AI Shutdown Resistance, Self-Replication & Ecology
Suggested citation: The Daily AI Digest, created by Alexander Panetta — dailyaidigest.net (May 25, 2026).