Trajectory-Based Liability: Regulating High-Capability Agentic AI in the 2026 Capability Frontier

The proliferation of high-capability agentic artificial intelligence (AI) systems in 2026 marks a fundamental shift from passive algorithmic tools to autonomous actors capable of executing irreversible real-world actions across multiple domains. Empirical safety auditing of OpenClaw (previously Clawdbot), a widely deployed self-hosted agent with broad tool access, reveals critical vulnerabilities including intent misunderstanding, prompt injection susceptibility, and cascading operational failures that existing product liability and negligence frameworks cannot adequately address. This article argues that the legal and regulatory architecture must evolve from static product-centric models to dynamic trajectory-based governance. Drawing upon the empirical evidence of agentic risk amplification identified in the OpenClaw safety audit and emerging constitutional governance models, this article proposes a comprehensive regulatory framework comprising: (1) a tiered classification system for High-Capability Agentic AI (HCAAI) triggering strict liability; (2) mandatory trajectory auditing and logging requirements; (3) capital charge thresholds akin to Basel III/IV financial resilience standards; and (4) a safe harbour provision for deployers adopting constitutional AI governance. By grounding these proposals in established tort principles, financial regulatory analogies, and the empirical reality of 2026 agentic capabilities, this article offers a concrete pathway for juridical adaptation to the agentic era.