Anthropic Loosens the Nuts: Commerce Beats Safety

On February 25, 2026, Anthropic quietly softened its core "Responsible Scaling Policy" (RSP). Previously, the developer promised to slam on the brakes or stop training models if they showed dangerous autonomous capabilities. Now the wording has become much more vague.

This is a strategic shift for the entire industry. Training frontier models requires billions in CapEx, and investors are not ready to sponsor "brakes". Competition with OpenAI and xAI dictates the rules: scaling laws and time-to-market have taken absolute priority over theoretical risks. Ethics have capitulated to economics, and Big Tech's self-regulation has finally proven its inadequacy.

Source: WSJ / TIME