An Analytical Study of Algorithmic Bias Mitigation Frameworks in Multi-State Personal Lines Insurance for Regulatory-Ready Artificial Intelligence Underwriting

In the United States, state laws and regulations, such as Colorado Senate Bill 21-169 and New York Insurance Circular Letter No. 7, limit the use of artificial intelligence (AI) in personal lines insurance pricing. This study formalizes and assesses Neurosymbolic Governance, an architectural approach that embeds probabilistic AI risk scores within deterministic regulatory guardrails so that pricing remains actuarially sound and compliant with state regulations. The framework was tested through a simulation-based architectural evaluation using 10,000 synthetic underwriting profiles. In the standalone configuration, the AI pipeline exhibited a mean proxy discrimination gap of 14.48% (95% CI: 13.95%-15.05%) across demographic segments. With the deterministic guardrail layer enforced, 1,976 of 10,000 transactions (19.8%) were intercepted and recalibrated, reducing the mean gap to 0.14% (95% CI: 0.04%-0.45%), a relative reduction of 98.6% (95% CI: 96.9%-99.7%). Mean per-transaction latency rose from 120 ms to 125 ms (+4.2%), a small and practically insignificant difference, and a structured decision log linking each decision to a specific regulatory rule was produced for 100% of transactions, supporting the explainability of AI-driven decisions. The results indicate that Neurosymbolic Governance is a promising architectural approach to the conflict between the speed of predictive technology and the conservatism of insurance law, enabling carriers to modernize their current cloud-based systems while introducing fairness safeguards into production machine learning. The framework still needs to be validated on real-world data.