Sanders and Schneier: AI's fears are really capitalism fears

A Tech Policy Press essay by Nathan Sanders and Bruce Schneier argues that the AI debate keeps mashing together two different problem sets: things the technology itself does badly, like context loss, confabulation, and sycophancy, and things a market structure does with that technology, like resource capture, monopoly, and labor cuts. They borrow Ted Chiang's 2021 observation, quoted in the piece, that "most fears about AI are best understood as fears about capitalism," and spend the essay pulling those two threads apart.

Their sharpest illustration is medical. Give a physician an AI assistant and, in the authors' words, "the AI could give a doctor more time to do the human parts of their job." Or the same tool could let the managers of a practice hand one doctor "five times the patients" and fire the other four. Which outcome you get "is not a question of technology. It's a question of market incentives." The capability is identical; the economic wrapper decides who benefits.

To show the choice is real, the essay contrasts three postures. Switzerland's Apertus model, they note, was trained "entirely on data validated to be licensed for use with AI (not stolen), on pre-existing public computing infrastructure, and using renewable hydropower." Chinese labs like DeepSeek and Qwen are shipping "smaller, more efficient, more affordable models" on commodity hardware, and often giving them away. US frontier developers, with OpenAI and Anthropic named in the frame, sit at the opposite end, running capital-heavy retraining cycles. Same underlying technology, three very different political economies.

The reform list is short and blunt. Sanders and Schneier want companies "forced to pay the energy and environmental costs" of AI development, profits "taxed adequately and redistributed," and antitrust laws "strongly enforced." Two AI experts we track circulated the piece on the day it ran, a small signal the framing is landing with policy-facing readers as much as with builders.

This is a framework argument, not a numbers-driven investigation. The essay does not model what a Swiss-style public compute stack would cost at frontier scale, does not name which antitrust theory would bite hardest on foundation-model consolidation, and does not weigh whether Apertus or the Chinese models actually match US frontier capability. For a strategist reading it, the useful takeaway is that "AI safety" and "AI political economy" are not the same file, and the largest incumbents benefit whenever the two get filed together.