PulseExploreJournal ClubDebatesTrendingResearchersJournals
Instagram
HomeExploreJournal ClubTrending
Synapse
⌘+K
Synapse
April 8, 20260 citationsOpen Access

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

View Full Paper
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
Share
View Full Paper

Cite This Study

Ana Margarida Ferreira Vila Chã Esteves (2026) studied this question.

synapsesocial.com/papers/69d5f14b74eaea4b11a7ae77https://doi.org/10.5281/zenodo.19433911
Ask AI
Helpful
Bookmark
Share
View Full Paper