Regulatory Intelligence: A Viability-First Paradigm for Artificial Intelligence

This work introduces Regulatory Intelligence (RI), a viability-first paradigm for artificial intelligence in which intelligence is defined by a system’s ability to maintain internal coherence, stability, and bounded operation under stress rather than by task-level accuracy or optimization performance. The paper formalizes RI as a control-theoretic and dynamical systems framework grounded in viability theory, homeostasis, and stability analysis. Intelligence is treated as a regulatory property: a system must remain within a viable state space while engaging with high-entropy, contradictory, or adversarial inputs. The paradigm explicitly rejects scale-driven learning and performance maximization as primary objectives. A concrete neurosymbolic implementation, SpiralBrain-v3.0, is presented as a reference instrument for studying regulatory cognition. The system operates over a fixed-dimensional cognitive manifold and employs geometric homeostasis, hazard-based regulation, and elastic (non-persistent) adaptation to preserve internal integrity. Adaptation occurs within bounded runs and resets between executions, enabling falsification of unintended learning or cross-run plasticity. The work emphasizes falsification-driven evaluation rather than benchmark competition. Stability, boundedness, recovery, and integrity are treated as first-class metrics, with empirical demonstrations showing sustained homeostasis under cognitive stressors. Lower benchmark scores are interpreted as intentional regulatory throttling rather than architectural failure, reflecting deliberate prioritization of viability over raw accuracy. This paper positions Regulatory Intelligence as a complementary alternative to optimization-centric AI paradigms and as a foundation for safety-critical, interpretable, and instrument-grade artificial cognitive systems.