Follow ZDNET: Add us as a favorite source On Google.
ZDNET Highlights
- AI can increase productivity and improve data access.
- Data concerns have caused tech leaders to pause the rollout.
- Long-forgotten insights emerge with AI signals.
Agentic and generative AI have opened up information and insights to professionals in enterprises. However, evidence suggests the trend may be too much of a good thing. At a recent conference, veterans of enterprise AI rollouts issued cautionary words to professionals considering plunging headfirst into AI.
The problems these professionals faced also led to a temporary halt in AI rollouts meant to boost employee productivity, as executives reassessed the amount of information that could have been exposed internally. At the same time, officials who spoke on a panel at the recent Veeam conference in New York City emphasized that AI was not the source of the challenge. Both panelists’ organizations had accumulated vast troves of data, and one needed a new governance structure.
Also: 96% of IT Professionals Now Use AI: Their Top 7 Agentic Applications and Biggest Implementation Barriers
Steve McIntyre, senior vice president at Fidelity Investments, described how his 400,000-employee company saw data long hidden in the gaps of his organization – for example, on SharePoint sites or in network-attached storage – suddenly surface through AI signals. “This was not an AI problem,” he said. “It was the productivity of AI and its ability to find things instantly.”
Wim Guerdon, chief architect of enterprise tech at EY, described his company’s challenge as reducing data ownership across its global network of independent associates – data that was also being exposed through its AI engine. “When the search for large enterprises began, all kinds of things started coming up in places where people went,” he said.
“EY Global doesn’t own any of the data. Each member firm has its own data. That’s where the questions were first raised. What is it all about? How many SharePoint sites? We had several petabytes of data, and it was the Wild West. There was no lifecycle management on these SharePoint sites, and half of them had no owner. We didn’t know when they were last accessed.”
Too: 51% of Professionals Say AI Workflow Reduces Their Productivity – Stop It in 2 Steps
At Fidelity, information was coming out of a vast library of PowerPoint and PDF reports. “We have the entire history, decades of Fidelity research notes, like PDFs,” McIntyre said. “We gave out some licenses for Copilot, and immediately, two days later, legal came to me and said we had an AI problem. One of my team did a search to find something and the AI came back with all those PowerPoints that were on SharePoint years ago.”
AI is a “tremendous search engine that runs at speed,” McIntyre added. “Suddenly, it’s finding everything it has access to, and exposing it to us in a meaningful way. Everyone thought we had an AI problem, but what it showed was the problem of securing data. That problem hit home when we immediately realized we had all this data we didn’t think we cared about – unstructured data – and along came LLM, and suddenly all that data became gold.”
installing handrail
At EY, as the doors to its vast data stores opened to AI, the priority was “to figure out who owns the data,” Geurden said. “The second thing we did was we shut everything down.” Users could access the CoPilot tool only if they had a license.
Too: Building an agentic AI strategy that pays off without the risk of business failure
The data ownership verification process involves identifying and labeling data found across the EY enterprise, Guerdon continued. For example, labels include “Confidential” or “Financial Services.”
Considering the challenge of human labeling with a 25% annual turnover rate, AI itself offered a means to help label the company’s knowledge store of unstructured data, Guerdon explained.
However, labeling needs to go deeper than simple high-level tags. “The first thing we need to know is what was there when the AI was running,” Geurdon said. “We need historical pictures, ed.” Then, “We need to go far beyond labeling confidential information. We need geo-restriction, geo-labeling, line-of-business labeling attached to our contracts, because we get massive amounts of customer data that specify what we can and can’t do.”
Too: Over 80% of US government agencies already use AI agents – and that’s only the beginning
All this metadata would have to be codified into contracts, he said: “That’s the easy part. Then we have to codify it into some technical structure. That is, for now, still very, very cumbersome.”
Officials stressed that governance is key to success in all aspects of these AI implementations. “We need to know what is being used,” McIntyre said.
“It brings up the idea of shadow AI, shadow IT, all those types of things – and this last point goes back to data. We need to know that the asset lists are accurate. Are they registered and aligned with approved use cases? That way, at least we know if someone is working on something, they should be using the cloud, because it’s tied to a particular project that was approved for it.”
Also: These 4 critical AI vulnerabilities are being exploited faster than defenders can respond
Further, “We have to think about what is the safe environment that we want these agents to run in?” McIntyre continued: “How do we want them to interact with fundamental models? What architecture do we put in place to funnel all that activity into one place that gives us the right visibility and telemetry so we can see whether the agents and applications using AI are behaving in the way that was intended? Or misbehaving?”
An additional challenge — perhaps the most vexing one for all digital leaders right now — is establishing agent identity, McIntyre said: “How do you give an agent identity? They then become an employee. But what if my agent only lives for a few seconds? It’s a really interesting problem, and I don’t know if anyone has solved it well yet.”
