Artificial intelligence agents were supposed to streamline business operations and cut costs.
Instead, many companies are discovering that these tools are returning unexpected data, sometimes sensitive, sometimes embarrassing.
“Four out of five companies said they were already aware of an instance where an AI request had returned unexpected information,” said Mark McClain, CEO of SailPoint, during a recent Fox Business interview.
AI That Gives Away Too Much
“These AI agents are designed to go out and find information to help the business accelerate, or do what it does better,” said McClain.
“But there are unfortunately plenty of instances of them returning unexpected information.”
He gave the example of someone simply looking up general pay data, only to receive a colleague’s W-2 form. That kind of leak isn’t just awkward, it’s a potential legal and security liability.
Even Salesforce CEO Marc Benioff flagged this issue earlier, expressing concern that Microsoft’s Copilot might be showing employee emails it shouldn’t.
McClain warned that bad actors could eventually find ways to exploit these AI systems to uncover information that should remain private.
“This is sort of an anecdote that feeds that concern,” McClain said, referring to a recent Microsoft case involving what the company called a “click-free compromise,” where an email alone triggered an AI agent to send sensitive information to the wrong person.
The Human Touch Isn’t Going Anywhere
While some businesses rushed to replace workers with AI tools, many are now realizing it was a mistake.
A Gartner survey found that half of executives plan to scrap their earlier intentions to shrink customer service teams by 2027.
“The human touch remains irreplaceable in many interactions,” said Kathy Ross, a senior director at Gartner.
“Organizations must balance technology with human empathy and understanding.”
That reality is setting in. Klarna, the finance startup that laid off 22% of its workforce in 2024 to make room for AI, reversed course in May and announced new hiring.
Not Ready For Prime Time
Despite the hype, current AI agents aren’t delivering as promised.
As of April, even the best agents could only complete about 24% of tasks given to them, according to internal data cited in a Futurism report.
Critics say companies bought into the buzz without understanding what AI agents actually do.
“These ‘agents’ are branded to sound like intelligent lifeforms that can make intelligent decisions,” wrote tech critic Ed Zitron.
“But are really just trumped-up automations that require enterprise customers to invest time programming them.”
In reality, only 45% of corporate IT managers even have formal AI policies in place, and most cite security and integration problems as their top concerns.
What It All Means
AI agents aren’t the plug-and-play solutions companies hoped for.
They need strict access controls, clear policies, and often, a lot more human oversight than initially expected. For now, the rush to automate everything is slowing down, and with good reason.
Some companies are even reversing course and rehiring employees they once replaced with AI.
It’s becoming clearer that AI works best when paired with experienced people who can guide its use, not as a full replacement.