Informal Leadership Structures and AI Safety
Influential networks at Anthropic, Open Philanthropy, and Constellation shape AI safety without official leadership roles.
A viral LessWrong essay by LawrenceC examines the leadership vacuum in today's AI safety movement. Following the FTX/SBF scandal that discredited some Effective Altruism (EA) figures, the community has moved away from recognized authorities like Peter Singer or Eliezer Yudkowsky. However, the absence of formal leadership hasn't eliminated hierarchy—it's created informal power structures centered around three key groups: Anthropic, whose research priorities and short AI timelines shape the field; Open Philanthropy/Coefficient Giving, which funds over 50% of AI safety work; and the Constellation network of researchers from Redwood Research, METR, and related organizations.
These groups wield significant influence despite not claiming leadership roles. Anthropic's public communications define safety discourse, while Open Philanthropy's funding decisions determine which projects survive. The Constellation network—originally just colleagues sharing office space—now functions as an informal elite through lunch conversations and elevator chats. The essay argues this creates transparency problems: influence flows through friendship networks rather than accountable structures, making it difficult for outsiders to understand how decisions get made or challenge prevailing views, even when those views come from well-intentioned experts doing valuable work.
- Anthropic shapes AI safety research priorities and public perception despite no formal leadership role
- Open Philanthropy/Coefficient Giving controls >50% of ecosystem funding, giving them disproportionate influence
- Constellation network (Redwood Research, METR, Anthropic) functions as informal elite through personal connections
Why It Matters
Informal power structures in AI safety could lead to groupthink and make the field less responsive to external criticism.