PRISM: A Pointer-Based Provenance Memory Architecture with Architectural Content Firewall for Goal-Coherent Multi-Agent Artificial Intelligence

Read full text Summarise with VIRAT
Authors: Srivatsav Gopinath
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Multi-agent AI systems suffer from goal drift, cascading errors, memory poisoning, and context window dependency. PRISM (Provenance-Routed Injection and State Memory) presents a four-layer memory architecture with an enforced Content Firewall that separates content-aware and content-blind components at the training and architecture level. Empirically validated across 33 autonomous retrieval optimization runs and 16 multi-agent coordination runs. Results: 100% goal coherence, 100% cascade containment, 100% adversarial resilience (vs TMCHT published baseline: 52.93% poisoning rate), 100% MEMTRACK quality (vs GPT-5 published baseline: 60% correctness ceiling). Patent pending USPTO March 20, 2026

Fields of Study & Subjects

Adversarial Robustness in Machine Learning Access Control and Trust Multi-Agent Systems and Negotiation

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
21 March 2026
Language
EN
Document type
Preprint