This paper introduces SIS-10 (Safety Integrity System–10), a minimal axiomatic kernel comprising ten formally independent axioms that collectively guarantee safety invariance in autonomous and AI-driven systems. Unlike probabilistic safety frameworks that quantify tolerable failure rates, SIS-10 defines safety as a logical invariant derived from first principles: no axiom is derivable from the remaining nine, each constrains a distinct behavioral dimension, and their conjunction is both necessary and sufficient for invariance, bounded recovery, and compositional safety. We formalize the system model with extended definitions for multi-step transitions and parallel composition, prove axiom independence constructively via finite-state countermodels, and establish nine theorems covering safety invariance, bounded recovery, timed progress, idempotent retry safety, compositionality preservation, non-absorbing unsafe states, observable degradation, fail-safe containment, and deployment safety. A Metatheorem on Kernel Tightness shows that each axiom is load-bearing for at least one theorem. We establish a novel LTL Encoding of all ten axioms and prove that SIS-10 compliance is decidable in PSPACE for finite-state systems. The kernel is instantiated on a class of event-driven transactional architectures with Apache Kafka as a concrete exemplar and operationalized through a formalized CI/CD safety gate and runtime monitoring framework. A five paradigm comparison positions SIS-10 relative to probabilistic SIL frameworks, formal verification, control-theoretic stability, process algebras, and emerging AI governance mandates, supported by a formal coverage table showing which axioms each paradigm addresses. SIS-10 does not replace existing safety standards; it defines the logical foundation they presuppose but have not formalized. Keywords: safety integrity, axiomatic kernel, formal methods, linear temporal logic, autonomous systems, CI/CD enforcement, functional safety, compositional verification, PSPACE decidability.
Building similarity graph...
Analyzing shared references across papers
Loading...
Usman Zafar
Building similarity graph...
Analyzing shared references across papers
Loading...
Usman Zafar (Fri,) studied this question.
synapsesocial.com/papers/69edad274a46254e215b4c91 — DOI: https://doi.org/10.5281/zenodo.19723514