Semantic indeterminacy in agentic negotiation: agreement after the recession of the principal

Abstract

Agentic AI systems increasingly do not merely respond within negotiations but participate in constituting them: they can add issues, instantiate participant agents, and modify procedural rules while an interaction is under way. Human action can be compositionally produced without thereby becoming authorless, and legal institutions routinely attach consequences to conduct without identifying an author of every contribution. Our claim is therefore limited: agentic generation can separate three relations that institutions often treat as aligned: causal production, intentional expression, and institutional attribution. We call the resulting condition semantic indeterminacy. An added element may result from several generative influences, including principal intention, agent-side optimization, training distributions, and interactional contingency, while information about its causal production does not, by itself, determine whose intention it expresses or how it is institutionally attributed. We show how this condition can appear across three sites of bargaining: issues, participants, and procedural rules. We then trace two implications. First, causal analysis alone does not determine who is responsible for a generated element or on what basis, although contract, agency, and product-liability doctrines can assign legal consequences under specified conditions. Second, as a limiting case, agentic interactions may occur in post-principal agentic spaces, where no human provides move-specific direction and no identifiable human or institution occupies a legally or institutionally recognized role. Existing governance frameworks remain grounded in identifiable human authority and accountability; their limits become visible where that ground is no longer reliably available. The paper provides a conceptual framework for identifying this residual problem rather than claiming to solve it.