AI Safety Needs Public Rules, Not Private Truces

Artificial-intelligence governance is often narrated as a contest between two simple options: let companies race, or ask the same companies to agree to slow down. Neither option is sufficient. Competition can produce innovation while rewarding speed over care. Private coordination can support shared safety work while concentrating rule-making among firms whose commercial and geopolitical interests are substantial.

The missing question is legitimacy. Who defines the danger? Who decides the acceptable pace? Whose evidence counts? Which learners, workers, churches and countries participate? Who can inspect compliance, challenge unequal effects and obtain remedy? A safety decision is not trustworthy merely because several powerful chief executives agree with it. Nor is an unsafe decision justified because competitors would otherwise move faster.

Responsible governance requires a public-interest chain: demonstrated risk → affected people → lawful authority → transparent rule → independent evidence → fair competition → institutional implementation → human appeal → review and revision. AI safety needs cooperation, but cooperation must be accountable to more than the companies being governed.

The Associated Press reported on 20 September that four paying subscribers had filed a proposed class action in the United States District Court for the Northern District of California against Anthropic, OpenAI, SpaceXAI and Google. The plaintiffs allege that public expressions of support for coordinated slowing of AI development amounted to an unlawful agreement that would restrict competition…