Rules or Character? Scaling Laws for AI Safety Design

2026-08-13

AI Summary

This study investigates how AI systems should dynamically balance rule-based safety mechanisms against behavior-shaping approaches as deployment scale increases, aiming to minimize both expected harm and tail risk. We formalize this trade-off for the first time, introducing a parameter α to represent the allocation of resources between the two strategies. Incorporating factors such as filter degradation, common-mode failures, and baseline behavioral vulnerability, we evaluate the trade-off using comparative statics, a multiplicative Pareto damage model, Monte Carlo simulations, and Conditional Value-at-Risk (CVaR). Our results show that the baseline vulnerability of behavior shaping is the dominant determinant of the optimal strategy. As deployment scale grows, the optimal α* shifts modestly to substantially toward behavior shaping (Δα* = +0.01 to +0.21), reaching offsets up to 0.50 under high vulnerability; moreover, at large scales, the CVaR-optimal and expected-harm-optimal solutions converge.

Abstract

Artificial Intelligence (AI) safety systems combine character shaping (e.g., Reinforcement Learning from Human Feedback [RLHF], Constitutional AI), which modifies behavioral distributions at training time, with rule enforcement (e.g., output filters, safety classifiers), which blocks harmful outputs at inference time, yet little formal analysis exists on how their optimal balance should change as deployment scales increase. We introduce a stylized comparative-statics model that parameterizes safety design as a resource allocation alpha in [0,1] between these two approaches, incorporating scale-dependent filter degradation, common-mode failures, and character fragility -- the risk that shaped behavior degrades or collapses under novel conditions. Under a multiplicative Pareto damage model, we derive closed-form expected harm and supplement it with tail-risk (CVaR) analysis via Monte Carlo simulation. Across three scenarios (optimistic, moderate, pessimistic), the optimal alpha* is interior or at the rules-only boundary and shifts weakly toward character shaping as deployment scale T grows, from negligible (Delta alpha* = +0.01) to pronounced (Delta alpha* = +0.21) depending on scenario. The dominant parameter is the baseline character fragility rate p^(0)_frag, which shifts alpha* by 0.50 across its range -- far exceeding the effect of tail severity, filter quality, or common-mode failure probability. CVaR and expected-harm optima converge at large T. These results suggest that safety architecture decisions depend less on deployment scale per se than on the reliability of character shaping under distributional shift.

Problem

• AI safety
• scaling laws
• character shaping
• rule enforcement
• distributional shift

Innovation

• AI safety
• scaling laws
• character shaping
• rule enforcement
• distributional shift