AI

Lineage-Aware Memory Governance: A Derivation-Gated Framework for Privacy-Preserving Column-Level Access Control in Enterprise AI Agents

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 08/10/2026, 16:18:24 UTC.

Briefings

This article proposes the Analytical Memory Unit (AMU), a new memory schema for enterprise AI agents that attaches a full derivation lineage graph to cached results. The AMU introduces a retrieval policy that blocks access to sensitive columns unless the requester is explicitly authorized for every column touched, preventing both data leakage and silent KPI conflicts. The authors prove by construction that this policy blocks unauthorized derived features, achieving O(n) worst-case performance. Experiments show that achieving 90% lineage completeness reduces cross-department leakage by 18.8-25.5% compared to naive content-gated systems, while maintaining 81.5-82.6% memory reuse with minimal overhead.

Key points

• The Analytical Memory Unit (AMU) attaches a full derivation lineage graph to every cached result in enterprise AI agents.
• A retrieval policy blocks retrieval of results derived from sensitive columns unless the requester is authorized for every column touched.

Context

Enterprise AI agents sharing a memory store face risks of sensitive data leaking through legitimate results and departments computing the same KPI through conflicting logic. Existing systems gate retrieval by content, ownership, or role but miss cached insights embedding forbidden columns. The AMU provides a practical governance layer for shared agent memory, supporting EU AI Act compliance.

Limitations

The study uses a feasibility demonstration with LLM-generated SQL rather than production viability evidence. Leakage reduction is based on recorded lineage completeness rather than measured leakage data. The 90% lineage completeness target is a conservative deployment target, not an empirical measurement. The O(n) worst-case performance is a conditional design guarantee, not an empirical claim.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 09/11/2026, 17:29:37 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeihxci2syzawevu2qsyhvv4jc2bhxrpt7uq3ojgwvomdmqwuq3kd7m