Summary
Following the midterm elections, Congressman Greg Casar posted “Ban RSI” on X, calling for a prohibition on recursive self-improvement, meaning AI systems that help build the next generation of AI. This video from David Shapiro examines the push, which also includes a Bernie Sanders-backed bill targeting superintelligence, and argues that a ban is both unlikely to pass and a bad idea.
Shapiro walks through five premises he considers flawed. He argues that RSI is not a single thing but a bundle of ordinary skills such as writing code and running experiments, citing figures like Claude writing more than 80% of the code Anthropic merges and OpenAI’s stated goal of an automated AI researcher by March 2028. He also points to Anthropic’s study of Claude-based automated alignment researchers, which outperformed 28 human safety researchers on several problems, as evidence that safety work is being automated too.
The second half makes the case that courts, customers, the military and investors already steer AI toward being safe and controllable. Examples include the Canadian airline chatbot tribunal ruling, Anthropic’s reported 55% share of business AI model spending, and the Pentagon dispute over control. Viewers interested in AI policy, regulation and the safety debate will find a detailed counterargument to the ban proposals.
📺 Source: David Shapiro · Published November 10, 2026
🏷️ Format: Opinion Editorial
