A Case for AI Safety via Law

Legal frameworks may be more practical than technical solutions alone. The paper argues that established legal systems—which have evolved over centuries to handle complex human behavior—offer a proven model for AI safety that goes beyond technical interventions like off-switches or isolation environments.

Law provides accountability mechanisms that other approaches lack. Unlike training systems to learn human values or relying on human intervention, legal systems create clear processes for enforcement, litigation, and consequences, establishing responsibility when AI systems cause harm.

Context-specific rules are essential for AI governance. The paper defines law as rules that apply to "particular agents in specified domains/contexts," suggesting that one-size-fits-all technical solutions may be insufficient and that AI regulation needs to be tailored to specific applications and industries.

Legal approaches address a gap in current AI safety research. The paper notes that law-based solutions have been historically overlooked in AI safety discussions (despite inspiration from Asimov's fictional laws of robotics), suggesting the field may benefit from incorporating legal and regulatory expertise alongside technical research.

Enforcement and management processes matter as much as rules themselves. The paper emphasizes that effective legal systems require not just prohibitions and prescriptions, but also processes for enacting, managing, and enforcing rules—highlighting that AI safety requires institutional structures, not just technical safeguards.