The question that haunted the 2018 Uber self-driving fatality in Tempe, Arizona—who bears responsibility when an AI system causes harm?—remains fundamentally unresolved across enterprises deploying generative AI at scale today. That accident prompted fundamental questions about liability chains: Was responsibility with the safety driver, the engineers who built the algorithms, executive leadership, or regulators who failed to enforce standards? Five years later, as companies from ServiceNow to Amazon deploy AI agents into mission-critical processes like employee onboarding and hiring decisions, organizations are beginning to construct internal accountability frameworks. Yet these remain largely proprietary and inconsistent, creating a patchwork where responsibility gets distributed rather than clearly assigned.

ServiceNow's approach offers a instructive case study. The company has embedded AI agents directly into business workflows while simultaneously appointing Jacqui Canney as chief people and AI enablement officer—a role specifically designed to govern how AI automates decisions affecting employees. This dual structure attempts to thread a needle: accelerate AI deployment while creating clear oversight. However, most enterprises lack such dedicated governance roles. The more common pattern mirrors what happened with Amazon and PwC's recent AI-driven layoffs: companies cite efficiency gains while responsibility diffuses across product teams, data scientists, and executives, making it nearly impossible to trace decisions back to accountable individuals when outcomes harm workers or customers.

The practical stakes are escalating as AI shifts from experimental tools to operational infrastructure. When generative AI compresses the cost of first attempts—generating code, analyses, and prototypes at near-zero marginal cost—companies push more decisions through automated systems. But accountability standards haven't evolved proportionally. Unlike aviation, where regulatory frameworks explicitly assign responsibility for system failures, AI deployment currently relies on vague principles and post-hoc investigations. Until industries establish clearer liability frameworks and companies designate unambiguous accountability owners for AI systems, the pattern established by Tempe will repeat: when AI systems cause measurable harm, everyone claims innocence, and no one bears consequences.