Title: From Control Boundaries to Insurance Claims: Rebuilding AI-Mediated Losses via the CER Framework

Abstract

When generative or agentic AI systems within an insured organization cause financial harm, traditional event reconstruction is insufficient. Because these systems continuously evolve their internal states through reasoning, data retrieval, tool invocation, and action execution, a state reconstruction approach is required. The critical inquiry extends beyond identifying the specific loss to determining the system’s permissible actions, its actual conduct, and whether this reconstructed narrative supports a viable insurance claim. This study examines scenarios where the insured’s AI system constitutes part of the causal chain, encompassing failures triggered by external factors such as prompt injection, retrieval-augmented generation (RAG) poisoning, malicious tool outputs, credential exploitation, and data poisoning.

To address AI residual risk transfer, this paper introduces the CER framework, a diagnostic tool designed for use-case-level analysis. The framework operates on three pillars: * C (Control Boundary): Evaluates whether the system operated within an enforceable operational envelope. * E (Evidence Reconstruction): Assesses whether retained artifacts allow for the reconstruction of the system’s state and the underlying causal chain. * R (Insurance Response): Determines if the reconstructed loss qualifies for coverage, verifying both market availability and the placement of policies for the insured, alongside the requisite proof for claim recovery.

The paper offers three primary contributions: it defines the unique challenges of AI-specific reconstruction, operationalizes this definition through the CER framework, and establishes the standard for claim-grade evidence in AI reconstruction. Real-world applications are illustrated through reported incidents involving PocketOS and Replit agentic database deletions, as well as the adjudicated case of Moffatt v. Air Canada, which serves as a precedent for output and reliance disputes.

Keywords: AI systems; CER framework; residual risk transfer; agentic AI; generative AI; AI insurance; evidence reconstruction.

Source: arXiv Generated at: 2026-06-03 00:00:00 UTC