What question did this study set out to answer?

The research aims to identify and solve a previously unrecognized safety vulnerability in write-ahead audit architectures.

February 19, 2026Open Access

The Zombie Intent Hazard: A Safety Violation in Write-Ahead Audit Architectures for Safety-Critical Systems

Key Points

The research aims to identify and solve a previously unrecognized safety vulnerability in write-ahead audit architectures.
Identified the zombie intent hazard in non-volatile memory systems.
Demonstrated the hazard's oversight by modern AI systems.
Developed a formal solution using session-bound verification and TLA+.
Confirmed that the zombie intent hazard allows execution of stale intents under changed conditions.
Proved the proposed solution's correctness through temporal logic analysis.

Abstract

This publication identifies and formally solves a previously undocumented safety vulnerability in write-ahead audit architectures: the Zombie Intent Hazard. This hazard occurs when a system persists an intent record to non-volatile memory, crashes before actuation, and upon reboot executes the stale intent despite changed environmental conditions. We demonstrate that this vulnerability was missed by three frontier AI systems (GPT-5.2, Claude Opus 4.5, Gemini 3) during independent architectural review. We present a formal solution using session-bound verification and prove its correctness using TLA+ temporal logic.

Read Full Paperexternally

Demander à l'IA

Bookmark

View Full Paper