2026年人工智能治理：背景、控制与企业规模

AI GOVERNANCE IN 2026:Context, Control, andEnterprise Scale Fern Halper, Ph.D. Contents AI Governance in 2026:Context, Control, and Enterprise Scale Market Context: AI Acceleration Meets Governance Reality3 Key Trends in AI Governance6 Trend 3: The Shift Toward Agentic AI Raises theGovernance Stakes............................8 Trend 4: Unified and Automated GovernanceBecomes Essential............................9 Governance as the Foundation for Scalable AI10 Context Gap, Platform Trap. . . . . . . . . . . . . . . . . . . . . .11Governance Provides Context and Control. . . . . . . . . . . .12Governance Everywhere. . . . . . . . . . . . . . . . . . . . . . . .12 About the Sponsor14 About the Author15 About TDWI Research15 Market Context: AI Acceleration MeetsGovernance Reality Artificial intelligence has entered a new phase of enterprise adoption.In TDWI research, generative AI now ranks as the top analytics priority,with agentic AI gaining traction as organizations look beyond contentgeneration toward task execution and automation. However, while many organizations view AI as a growth opportunity,relatively few have established mature governance frameworksto manage it. In a late 2025 TDWI survey, only about 26 percent oforganizations considered their AI governance mature, compared to31 percent for data governance (Figure 1).1Additionally, nearly halfdescribe their AI governance as immature. At the same time, just 36percent believe their data foundation is ready for AI (not shown). This maturity gap is emerging at a time when AI systems arebecoming more embedded, more autonomous, and arguably moreconsequential. Regulatory scrutiny is increasing globally. AI capabilitiesare being introduced through third-party tools and enterprise softwareupgrades, sometimes without centralized oversight. Unstructureddata use, i.e., documents, contracts, policies, emails, transcripts, and AI-generated content, is expanding rapidly and now plays a central role ingenerative and agentic AI use cases. Historically, structured data governance and information governanceoperated in parallel. Today, generative and agentic AI systems relyheavily on unstructured data. This introduces new governancerequirements: document validation, version control, lineage forunstructured assets, access control at retrieval time, monitoring ofupstream changes, and evaluation of faithfulness and relevance in AI-generated responses. The data in Figure 1 illustrates that AI governance lags datagovernance, and even data governance maturity remains uneven.At TDWI, we see many organizations still struggling with core issuessuch as data silos, unclear data ownership, lack of standardization,and inaccurate data. AI governance introduces additional layers ofcomplexity. The top challenges cited by respondents in a recent survey(Figure 2) include lack of clear guidelines and policies (51%), lack ofskilled personnel to manage AI models (50%), ethical concerns suchas bias and transparency (36%), and lack of controls over generative AIleading to shadow AI (32%).2 What challenges have you encountered withimplementing AI governance in your organization? These findings highlight an important point. AI governance requiresnew competencies that extend beyond traditional data governance.Whereas data governance focuses on ensuring that data is accurate,consistent, timely, and appropriately protected, AI governance extendsthose principles to systems that learn, generate, reason, and act. Itaddresses questions such as: …agentic AIintroducesnew technical,organizational,and governancechallenges.” •Is this model aligned with business objectives and company policy?•Are all AI assets fully inventoried and documented?•Is metadata (ownership, purpose, risk level, usage) clearly definedand maintained?•Do we have end-to-end traceability of how AI outputs areproduced?•Are the inputs to these models sound?•Can the output be explained? •Is the model degrading or drifting?•Is the model (or the agents) hallucinating?•What happens if the system acts incorrectly?•How should agents interact with one another? AI governance spans policy, technical controls, life cycle management,and risk mitigation. It requires coordination across businessstakeholders, data teams, engineering, legal, compliance, and security. While this seems like a lot to consider, the reality is that governancecan be an enabler, if performed properly. In TDWI research, we seethat governance maturity correlates strongly with measurable impact.Organizations with stronger data governance report higher levelsof efficiency, better data quality, improved compliance, and fasterinsights. They are more likely to implement observability, lineage, andmonitoring tools, and more likely to report top- or bottom-line benefits.In contrast, weak governance is associated with inconsistent practices,compliance exposure, fragmented ecosystems, and erosion of trust. Rather than slowing innovation, governanceprovides gua