Enterprise AI Rollouts Halt as 400,000-Employee Fidelity, EY Expose Hidden Data Risks
Updated
Updated · ZDNet · May 28
Enterprise AI Rollouts Halt as 400,000-Employee Fidelity, EY Expose Hidden Data Risks
3 articles · Updated · ZDNet · May 28
Temporary rollout halts hit enterprise AI programs after internal copilots and search tools surfaced long-forgotten files, reports and SharePoint content that executives had not expected employees to retrieve so easily.
Fidelity said the issue appeared within 2 days of issuing a few Copilot licenses, when legal flagged old PowerPoints and PDFs suddenly returned by AI across its 400,000-employee organization.
EY responded by shutting broad access and limiting Copilot to licensed users while it traced ownership across multiple petabytes of data, much of it in SharePoint sites with no clear owner or lifecycle management.
Both companies said the root problem was not AI itself but weak data governance, pushing efforts to label unstructured data, track historical versions, enforce geo and contract restrictions, and curb shadow AI.
The episode underscores a wider enterprise challenge: generative and agentic AI can unlock productivity only if companies first know what data exists, who owns it, and how AI agents should be identified and monitored.
With AI agents now outnumbering staff 82:1, who is truly in control of your company's most sensitive data?
Your new AI assistant just unearthed a ten-year-old legal liability. Is the AI to blame, or your data governance?
When AI Hallucinates: The EY Report Retraction and the 30% Growth Challenge Facing Enterprise AI Integrity
Overview
In May 2026, EY retracted a major AI-generated study after researchers discovered instances of AI hallucinations in its content. This incident highlights the recurring risk of relying on AI for information without strong human verification, as firms continue to face challenges with inaccurate or fabricated data. The retraction serves as a wake-up call for the entire enterprise AI landscape, emphasizing the urgent need for rigorous oversight and comprehensive strategies to validate AI outputs. As more consulting firms embrace AI, this event raises important questions about how often such errors will occur before robust verification mechanisms become standard practice.