Improper Validation of Identifier Uniqueness After Lossy Normalization: A Cross-Framework Weakness Class in AI Agent Orchestration Systems

This paper formally proposes a new software weakness class — Improper Validation of Identifier Uniqueness After Lossy Normalization (parent: CWE-697, Incorrect Comparison) — and substantiates it with two independently discovered, live-reproduced vulnerabilities in unrelated open-source AI agent orchestration frameworks: n8n and CrewAI. In both cases, a normalization step intended to standardize identifiers (URL path segments in one case, tool names in the other) discards information necessary to distinguish genuinely distinct identifiers, allowing a lower-trust component to silently displace a higher-trust one under identifier collision. We demonstrate that this is not an implementation-specific bug but a recurring architectural pattern, and provide a live, end-to-end exploit chain showing real-world consequence: an unauthenticated actor causing a fraudulent payroll change to be approved by an automated agent instead of the intended human-approval-only pathway.