Access Controls Will Solve the Dual-Use Dilemma

Evžen Wybitul

Keywords: access control, AI safety, modularization, gradient routing

Abstract: AI safety systems face the dual-use dilemma: it can be unclear whether to refuse certain requests, since they could be either harmless or harmful depending on who made them and why. Determining this requires examining their real-world context, but current safety systems cannot access this contextual information. Instead, they make arbitrary decisions that end up hurting both utility and safety: they sometimes refuse legitimate queries and other times fail to refuse harmful ones. To address this, we propose a conceptual framework based on access controls in which only verified users can access dual-use outputs. We describe the framework's components, analyse its feasibility, and explain how it addresses both over-refusals and under-refusals. While only a high-level proposal, our work takes the first step toward enabling more nuanced safety decisions: with better tools for managing dual-use content, model providers could enable users to access more capabilities without sacrificing safety, and give regulators new options for more targeted policies.

Submission Number: 73