Purpose-Bounded Context Exchange in Enterprise Agentic AI Systems: An Inter-Module Information Contract Architecture

Enterprise agentic AI systems exchange contextual information across functional modules, creating governance challenges when information crosses operational and organizational boundaries. This study distinguishes organizational governance boundaries from information exchange boundaries and proposes an Inter-Module Information Contract (IMIC) architecture for purpose-bounded context exchange. IMIC evaluates provenance, semantic and temporal validity, permitted information scope, and applicable originating constraints before downstream execution, independently of language-model reasoning. We compare IMIC with full-context sharing, reduced-context sharing, and strengthened structured handoff. In a deterministic benchmark of 32 synthetic conditions, joint correctness is achieved in 3, 10, 28, and 32 cases, respectively. The four differences between strengthened structured handoff and IMIC arise from conditions testing onward-transfer permission, recipient, purpose, and field scope. A downstream integration experiment evaluates 14 selected conditions across four local language models and three seeds, yielding 168 case–model–seed blocks and 672 method-level records. IMIC achieves correct exchange actions and information boundaries in all 168 blocks and joint correctness in 165 across four local LLMs and three seeds. Three permitted executions produce incorrect downstream responses despite correctly governed exchanges, demonstrating that information-exchange correctness does not guarantee correct downstream task performance. These results support the functional feasibility of the implemented originating-contract inheritance mechanism under controlled synthetic conditions. Generalizability to independent enterprise workflows, richer policy structures, multiple information providers, and derived information requires further validation.