No One to Blame: A Framework of Constitutive AI Unaccountability

Some AI accountability gaps cannot be fixed through better standards or transparency alone. The paper argues that certain structural configurations of AI systems, actors, and institutions make accountability "conceptually unachievable" rather than merely difficult—a fundamental shift from the assumption that accountability problems are solvable with enough effort and reform.

Open-source AI systems can create accountability voids where no one is responsible. When researchers applied the framework to OpenClaw (an open-source agentic AI system), they found an "inverted anthropomorphism" problem where the AI agent itself was the only clearly identifiable actor, making it impossible to assign responsibility to any human or organization.

Accountability problems stem from interconnected structural, technological, and normative issues. The study identified 20 distinct themes of unaccountability organized across three clusters (structural, technological, and normative) that reinforce each other through eight interdependencies, suggesting that fixing one problem in isolation won't solve the broader accountability challenge.

The framework provides a diagnostic tool for identifying when accountability is genuinely impossible. Rather than offering solutions, the paper operationalizes its findings as 20 diagnostic questions that can help organizations and policymakers recognize when they're facing constitutive unaccountability versus solvable accountability gaps—an important distinction for setting realistic expectations.

Current accountability thinking may be fundamentally misaligned with how autonomous AI systems actually work. The research suggests that traditional accountability mechanisms designed for human actors and hierarchical organizations may be inadequate for distributed, autonomous systems, requiring a conceptual rethinking rather than incremental improvements.