人工智能状况：拨开炒作迷雾，驾驭信任、安全与价值

Go Beyond theHype to NavigateTrust, Securityand Value Table of Contents Introduction4 Current state of Enterprise AI9 Signs of strengthening guardrails for AI17 Implementation status and impediments22 Addressing data security and data management issues withgenerative AI26 Information management strategies31 That’s a whole lot of data Data quality concerns with generative AI hinder value43 Capturing a return on AI investments47 Collecting feedback on AI tools from employees52 Interventions are strengthening AI literacy55 Recommendations Conclusion60 Introduction INTRODUCTION AI is no longer a future disruptor – it’s a present reality.But as adoption accelerates, trust in AI outputs is eroding. Organizations realize that without stronggovernance, resilient data strategies, and a commitment to quality, AI can just as easily become aliability as a competitive advantage. According to data from 775 global business leaders across financialservices, government, and healthcare, AvePoint found: AI usage fuels data security incidents. #1 75.1% of organizations reported at least one data security incident whereoversharing sensitive information negatively impacted them. As AI becomes more popular, scrutiny increases. #2 Microsoft 365 Copilot has become the top sanctioned generative AI tool, and over50% of employees have integrated the use of generative AI tools into daily andweekly work. Still, 88.3% of organizations have only rolled this out to some em-ployees, with pilot programs focused on data security and readiness. Security and accuracy concerns stall AIrollouts up to a year. #3 The two main reasons AI is not more widespread are inaccurate output (68.7%)and data security concerns (68.5%). These challenges are delaying AI rollouts upto 12 months for over three-quarters of organizations. Traditional information management frameworksfall short in the AI era. #4 While 90.6% of organizations have what they perceive to be an effectiveinformation management program or framework in place, few have dataclassification and incident prevention in place. Their overconfidence will createchallenges as data increases at a rapid clip. Data growth isn’t just a scale issue – it’s a security risk. #5 79.2% of organizations now manage 1 PB or more of data, up 25% from last year, andnearly 20% of organizations believe that in 12 months more than half of their data willbe created by generative AI. As data sprawls across cloud platforms, organizationsmust deploy true information lifecycle management to reduce their risk. Inaccurate AI output threatens rollouts anderodes employee judgment. #6 68.7% of organizations have slowed the rollout of generative AI assistants due toinaccurate AI output, and more than two-thirds of respondents express significantconcern for the impact of inaccurate outputs on employee judgment. Cultivating AI literacy is critical to the successof any AI rollout. #7 99.5% of organizations have used a range of interventions to strengthen AIliteracy among employees. Public AI use is finally being governed. #8 AI Acceptable Use policies have increased by 79.8% year-over-year, with 84.5% oforganizations now enforcing or developing clear AI usage guidelines. AI success is measured by customer impact – butthat’s where most fall short. #9 Enhancing customer insights and personalization is what organizations want toimprove when implementing AI, yet there is a 5.8% gap between what they hopeto achieve and what they actually do. AI adoption is measured by usage – not by impact. #10 89.6% of organizations use built-in or third-party reports to track how generativeAI assistants are used and how well they perform. But far fewer look at the humanor business impact using interviews or other qualitative methods. Data & Methodology: Key findings and data in this report comes from a 2025 global study by Osterman Research forAvePoint. Building onAvePoint’s 2024 AI & Information Management Report, this 2025 studycontrasts year-over-year data, revealing how enterprises are moving from AI experimentation toenterprise-wide enablement, emphasizing governance, risk mitigation, and measurable outcomes.775 respondents with responsibility for information management, data security, or AI programs attheir organization were surveyed. Current State ofEnterprise AI CURRENT STATEOF ENTERPRISE AI Enterprise AI adoption is accelerating Over the past year, more employees have gained access to approved generative AI tools and mostnow use them daily or weekly. Microsoft 365 Copilot and ChatGPT are used at similar rates, with only a 0.4% difference. However,Copilot has seen faster growth – rising from 47% to surpass ChatGPT, which grew slightly from 65%.Approval for Google Gemini also increased, growing faster than ChatGPT but slower than Copilot.Most organizations approve two (44.3%) or three (25.1%) generative AI tools for employee use.See Figure 2.1. Figure 2.1 Top generative AI tools sanctioned f