Agentic AI and Large Language Models: Responsible and Governed Intelligent Systems Across the Data Lifecycle

Big data governance has traditionally focused on data quality, access, privacy, security, lineage, provenance, and lifecycle management, while responsible AI has emphasized fairness, transparency, accountability, safety, and human oversight. Agentic AI brings these areas together because system behaviour emerges from the interaction of data sources, models, prompts, memory, tools, users, environments, and other agents. Failures may therefore originate not only in the model, but also in the orchestration layer, tool use, inter-agent communication, or the governance process itself. Model-level evaluation alone is insufficient. Governance must extend across the full data–AI lifecycle and combine design-time principles, organizational and policy mechanisms, runtime controls, human intervention, and post-deployment assurance.

Large language models are evolving from passive content-generation tools into agentic and multi-agent systems that retrieve, integrate, analyse, and act on data across complex environments. Their capabilities for planning, tool use, memory, autonomous decision-making, and inter-agent coordination create governance challenges that extend beyond model-level assessment. Risks may emerge at any stage of the data–AI lifecycle, including data acquisition, preparation, model development, agent orchestration, deployment, monitoring, incident response, and retirement. This Research Topic aims to advance conceptual, policy, methodological, and technical approaches for governing LLM-based agents across this lifecycle in different domains. It seeks contributions that clarify governance requirements, propose practical frameworks, develop assurance and oversight mechanisms, evaluate system behaviour, or provide evidence and lessons from implementation and deployment.

This Research Topic welcomes Original Research, Methods, Technology and Code, Systematic Reviews, Reviews, Policy and Practice Reviews, Perspectives, Hypothesis and Theory articles, Conceptual Analyses, Community Case Studies, and Brief Research Reports, subject to section availability. It focuses on the governance of LLM-based agents and multi-agent systems as data-intensive systems operating across the data lifecycle. Relevant themes include lifecycle-aware architectures for agentic data acquisition, integration, memory, and tool use; data quality, provenance, lineage, access, and documentation for agent operations; governance middleware, policy-as-code, and runtime controls; safety and security of agentic workflows, including prompt injection, tool misuse, and cascading multi-agent failures; behavioural traceability, action provenance, auditability, and assurance; evaluation of autonomy, controllability, reliability, and explanation reliability; and human oversight, escalation, accountability, and incident response. Conceptual frameworks, policy studies, technical methods, evaluations, and deployment case studies are welcome when they advance big-data management, data science, or machine learning and AI.