Anthropic's AI safety focus misses decentralizing reality, critic argues

The piece builds on Langdon Winner's 1980 essay "Do Artifacts Have Politics?" to argue that technology embeds and enforces political norms—a framework the author applies to contest Anthropic's structural analysis of AI. Where Anthropic sees AI as inherently centralizing (due to capital requirements, server-based inference, design choices by small teams, and token scarcity pricing out the developing world), the author identifies a countervailing force: the collapse of content moderation.

The author's core claim is that AI's ability to generate undetectable automated content has made traditional gatekeeping impossible. CAPTCHAs can no longer distinguish humans from machines; anything hard enough to filter AI also blocks legitimate users. This uncontrollable proliferation of bot spam has forced websites offline, shifting power away from institutions and toward individuals who can deploy automated tools. In this reading, Anthropic's regulatory focus on open-source models appears misaligned with the actual problem.

The critique also highlights a structural tension in Anthropic's position: the company depends on a business model where it controls frontier AI and restricts access, yet market forces push it to make tokens as widely available as possible. The author suggests this contradiction—combined with the fact that Claude already enables the harms Anthropic claims to prevent—makes the company's push for open-source regulation appear motivated by competitive advantage rather than genuine safety concern. The piece does not claim AI has no centralizing features, but argues that decentralization through bot spam and automated abuse is the dominant force shaping how AI actually affects institutions and individuals today.