AI Deployment Accountability Engineering: A Vision for Accountable AI in Safety-Critical Socio-Technical Systems
| Source: arXiv AI
Tags: AI governance, accountability, AI safety, agentic AI, compliance, socio-technical systems
A vision paper from ACM AI Summit 26 proposes AI Deployment Accountability Engineering (ADAE) as a new engineering subdiscipline — treating post-deployment accountability as a measurable system property rather than a model characteristic evaluated before launch.
Details
Current AI evaluation practice is predominantly model-centric: accuracy, robustness, fairness, and interpretability are assessed before deployment. But once an AI system operates inside a socio-technical environment with distribution shifts, institutional constraints, human feedback loops, and interactions among multiple AI agents, those pre-deployment properties are necessary but insufficient. ADAE (AI Deployment Accountability Engineering) proposes treating accountability as a deployment-layer property. The vision paper, presented at ACM AI Summit 26, organizes a research agenda around four pillars: structured discovery of context-dependent failure modes, privacy-preserving accountability measurement, system-level risk analysis for agentic AI, and translation of technical failures into operational and institutional risks. The paper is a vision piece — it articulates a research agenda and architectural goals rather than presenting empirical results. For enterprises deploying AI in healthcare, finance, or public services, ADAE names and frames a set of engineering problems that currently lack systematic tooling. For regulators, it offers a technical foundation for continuous compliance monitoring rather than point-in-time audits.