构建可信AI：数据平台核心要素

resilience and observability capabilitiesunlock AI innovation at scale TABLE OF CONTENTS Introduction: 6 Critical Platform Elements for Building Trusted AI3Governance5Security7Privacy10Interoperability12Business Continuity and Disaster Recovery13Observability16Make Trust an Enabler, Not a Roadblock18 INTRODUCTION:6 CRITICAL PLATFORM ELEMENTSFOR BUILDING TRUSTED AI not fragmented toolsThe rise of AI exacerbates longstanding challenges organization is built — and with the rise of AI, it’sonly becoming more important.Data is proliferating across more sources, governance challengesWhich of the following challenges does your in protecting and managing data, according toa2024 MIT Technology Review Insights Report.Data governance, security or privacy were citedby 59% of respondents as their top issue whendeploying AI at scale.These concerns are now compounded by new, organization face when it comes to deployingAI at scale?Data governance, security or privacy applications and clouds, in multiple formats.The advent of generative AI and large languagemodels (LLMs) means massive datasets — oftenincluding sensitive data like personally identifiableinformation (PII) and confidential medical andfinancial details — are now at the heart ofbusiness operations. To harness powerful AIagents and applications, companies need adata platform that doesn’t just help them unify,protect, analyze and share their data, but alsohelps safeguard it from unauthorized access,supports its compliance with regulations andmaintains resilience against disruption. emerging needs. To protect the massive, ever-growing datasets that fuel AI from outages andattacks, companies need robust business continuityand disaster recovery. And to understand andtroubleshoot the increasingly complex “black-box”nature of AI systems and their underlying datapipelines, built-in observability is a must. 49% Today’s environment demands a unified data platform that’s both powerful andeasy to use, deeply connected and fundamentally trusted. As organizations rushto deploy AI at scale, the stakes for getting trust right have never been higher.This requires a cohesive approach that integrates six critical elements from theground up:4.Interoperability:AI workloads increasingly span multiple clouds, regions and5.Business continuity and disaster recovery:In a world where your customers platforms — from training environments to production inference systems.Governing AI assets as they move between different environments requiresseamless interoperability to maintain consistent security policies, data lineageand model governance across your entire ecosystem. won’t tolerate downtime, a solid BCDR strategy is essential for maintainingoperational resilience. A unified platform helps safeguard your data estate,account metadata and governance policies across regions and clouds, ensuringyou can maintain operations and manage compliance demands even duringmajor outages or ransomware attacks.6.Observability:Monitoring black-box AI systems and increasingly complex models consume massive datasets that often include sensitive information,and data-centric regulations are increasing worldwide. This means companiesthat develop mature, integrated data governance capabilities are betterpositioned to deploy AI securely, mitigate regulatory risks and unlock AI’sbusiness potential.2.Security:AI systems present new attack vectors that traditional security data infrastructure and applications with traditional approaches oftenrequires weeks of configuration, leaving blind spots and delaying time to value.An effective approach to observability depends on fast, built-in capabilitiesthat allows organizations to demystify AI, accelerate troubleshooting andenjoy comprehensive visibility and insights.This guide dives deep into these core tenets of a trusted data platform and how approaches weren’t designed to handle. Organizations must implementprocesses and controls that protect not just data at rest, but also the AImodels, training datasets and inference pipelines from unauthorized accessand adversarial attacks. They also need a deep understanding of what’s intheir data so they can grant appropriate access and collaborate effectivelywhile preserving privacy.3.Privacy:AI’s hunger for data creates unprecedented privacy challenges. AImodels can inadvertently memorize and expose sensitive information from an integrated approach enables organizations to deploy AI at scale with moreconfidence and fewer compromises. training data. An ideal platform helps you harness AI’s power over sensitivedata while enforcing stringent privacy standards. GOVERNANCE AI is only as good as the data behind it — which makes data governance a high-stakes strategy. An idealgovernance framework preserves both the integrity and security of sensitive and nonsensitive data andsupports compliance with internal audits and regulatory laws around data residency and privacy.More trustworthy data A data governance framewo