What question did this study set out to answer?

The aim is to create a system that ensures AI decisions adhere to predetermined constraints before execution.

April 8, 20260 citationsOpen Access

Decision Safety System: Pre-Execution Enforcement of AI Decision Constraints

AEAna Margarida Ferreira Vila Chã Esteves

Key Points

The aim is to create a system that ensures AI decisions adhere to predetermined constraints before execution.
Developed a deterministic pre-execution enforcement layer for AI decisions.
Evaluated decisions against a formal set of constraints before execution.
Verified three formal invariants using TLA+ model checking.
Invalid decisions are blocked entirely and not returned at all.
Demonstrated effectiveness with a lending scenario that produces an audit record.

Abstract

A deterministic pre-execution enforcement layer for AI decisions in consequential domains. Evaluates every decision against a formal constraint set before execution. Invalid decisions are blocked and never returned. Three formal invariants verified using TLA+ model checking. Demonstrated with a lending scenario producing a verifiable audit record. Reference implementation of Article VII of the Constitutional Protocol for AI and Technology.

Ask AI

Helpful

Bookmark

View Full Paper

Cite This Study

Ana Margarida Ferreira Vila Chã Esteves (Mon,) studied this question.

synapsesocial.com/papers/69d5f14b74eaea4b11a7ae77 https://doi.org/https://doi.org/10.5281/zenodo.19433911