The AI Daily Brief: Artificial Intelligence News and Analysis

Episode summaries

1. The Case for an AI Token Tax

   May 28, 202622m

   A careful look at proposals to tax AI 'tokens'—why lawmakers and entrepreneurs are considering a usage levy, what it would try to solve, and the practical objections that make it hard to get right.

   The host surveys a rising policy conversation: politicians (Elizabeth Warren, Mallory McMorrow) and tech figures (Mark Cuban, Gabriel Weinberg, Dario Amade) have proposed levies on AI usage—often framed as per-token fees or excises on data centers—to capture revenue as work shifts from humans to agentic systems. He lays out a first-principles case: tax the locus of productive capacity, and tokens are an administrable proxy because providers already meter them. Then he runs through concrete counterarguments from economists and academics: tokens are a noisy proxy for value, tokenizer designs vary across languages and formats, per-token prices are collapsing, and a flat provider tax risks offshoring, distorted investment, and entrenching incumbents. The episode ends by favoring open engagement with novel policy while urging more targeted alternatives—consumption taxes, data-center community

   Key takeaways

   • — Elizabeth Warren (Time op‑ed) proposes taxing AI directly, starting with an excise on data‑center energy—"the bigger the data center, the more they pay."
   • — Mallory McMorrow’s policy calls for "a fraction of a cent charge per token" as a funding stream for worker retraining and apprenticeship programs.
   • — Mark Cuban suggested "less than 50¢ per million tokens," predicting ~ $10,000,000,000/year initially and potential 30x–100x growth over a decade if usage expands.
   • — Gabriel Weinberg (DuckDuckGo) proposed a 10% surcharge on token charges, roughly matching the employer payroll tax, and advocated locking proceeds into a displaced‑worker fund.
   • — OECD average tax on a single worker was 35.1% of labor costs in 2025; the IMF warned in 2024 that labor substitution could erode the income tax base if capital is taxed less than labor.
   Read the full transcript →

2. What the Pope Actually Said About AI

   May 26, 202626m

   A quick news roundup (Mythos, DeepSeek, Grok) followed by a close reading of Pope Leo XIV’s encyclical Magnifica Humanitas and its central plea to keep living humans — not market metrics or machine benchmarks — as the measure of success with AI.

   The episode opens with headlines: Anthropic’s Project Glasswing reportedly found more than 10,000 high‑ or critical‑severity software vulnerabilities, Mythos helped stop a $1,500,000 wire transfer, DeepSeek locked in a deep discount and pursues a $10,000,000,000 fundraise, and xAI announced Grok V9 (1.5 trillion parameters). The main segment parses Pope Leo XIV’s first encyclical, Magnifica Humanitas, placing it as a theological and civic intervention rather than a technical prescription. The host argues the encyclical’s core claim is simple and persistent: human dignity must remain the barometer for technology, and AI should not be conflated with human personhood. The episode closes by urging listeners to read the document themselves instead of relying on bite‑size social media takes.

   Key takeaways

   • — Anthropic’s Project Glasswing told partners it identified “more than 10,000” high‑ or critical‑severity vulnerabilities; testing on open‑source repos returned under a 10% false positive rate.
   • — Anthropic provided Mythos to roughly 50 trusted partners during the Glasswing rollout to focus on cybersecurity implications and disclosure cadence.
   • — Mozilla reported fixing 271 vulnerabilities found using Anthropic’s tools — reportedly more than ten times what they found with Opus 4.6 — and Palo Alto Networks shipped five times as many patches in a recent release.
   • — Anonymized reporting credited Mythos with detecting and preventing a $1,500,000 wire transfer to a compromised account.
   • — The White House approved a secret $9,000,000,000 budget request to let intelligence agencies build inference clusters (noted purpose: buy NVIDIA Blackwell chips and infrastructure).
   Read the full transcript →