The discourse surrounding Anthropic's approach to artificial general intelligence (AGI) development reflects two distinct philosophical positions regarding trustworthiness and safety governance. The tension between “only we can be trusted” and “no one can be fully trusted” represents fundamental disagreements about organizational capacity, systemic risk, and the appropriate distribution of AGI development authority.
The “only we can be trusted” positioning asserts that Anthropic, through its governance structures and safety-focused methodology, possesses superior institutional capacity to responsibly develop advanced AI systems 1). This view emphasizes Anthropic's constitutional AI framework, commitment to interpretability research, and stated prioritization of safety over capability advancement.
Conversely, the “no one can be fully trusted” perspective—reportedly held by a majority within Anthropic's governance discussions—reflects a more epistemically humble approach to organizational trustworthiness 2). This position acknowledges that even organizations with strong safety commitments face inherent limitations, conflicts of interest, and unpredictable systemic pressures when managing transformative technologies.
These positioning differences carry substantial implications for Anthropic's external communication and policy advocacy. The “only we can be trusted” framing, if adopted externally, could support arguments for concentrated development authority or favorable regulatory treatment of Anthropic relative to competitors. Conversely, “no one can be fully trusted” implies support for distributed oversight mechanisms, external governance structures, and systematic skepticism toward any single organization's self-governance claims.
The distinction becomes particularly acute when considering policy positions on AGI development coordination, international governance frameworks, and regulatory approaches 3). Organizations advocating “no one can be trusted” positions typically support transparent auditing, third-party safety evaluation, and institutional checks on unilateral decision-making—positions that constrain but also potentially legitimize their own development activities.
The coexistence of these viewpoints within Anthropic's governance discussions reflects tensions between stakeholder groups with different perspectives on organizational capacity and institutional humility. Safety researchers may emphasize the “no one can be trusted” framework as a methodological starting point, while leadership involved in capability development or business strategy may emphasize Anthropic's differentiated safety commitments and trustworthiness.
These internal debates shape organizational decisions regarding transparency, collaboration with other AI labs, support for external governance mechanisms, and communication with policymakers and the public 4). The resolution of these tensions influences whether Anthropic's public positioning emphasizes its unique safety capabilities or advocates for systematic institutional constraints applicable across the AI industry.
This positioning debate is not unique to Anthropic, reflecting broader tensions in safety-focused technology organizations between demonstrating organizational excellence and avoiding hubris regarding institutional capacity. Other advanced AI companies face similar questions about the relationship between technical safety work, organizational governance claims, and support for external oversight structures 5). The specific framing adopted by any organization signals its position on the legitimacy of concentrated versus distributed governance approaches to transformative AI development.