AI Agents in the Workplace: Why CIOs Are Losing Control of Their Own Tools

AI agents are multiplying faster than IT departments can keep up with them. According to a global study conducted by The Harris Poll for Dataiku among 685 chief information officers (CIOs), published in early November 2026, 81% of CIOs admit they no longer control the oversight of their AI agents deployed outside systems approved by IT.

Adoption outpacing governance

The finding goes beyond a technical anecdote. In most large companies, business teams are now creating their own AI agents to automate tasks (customer response, reporting, monitoring, procurement) without necessarily informing the IT department. This trend, often called “shadow AI,” accelerates short-term productivity gains but creates a blind spot for the teams responsible for security, compliance, and data quality.

What the Dataiku study reveals

• 84% of CIOs believe employees are creating AI agents faster than IT can govern them.
• 81% exercise no control over agents created outside approved systems or that go undeclared.
• 67% estimate that at least 51 AI agents are already running in production at their company.
• 83% have no standardized framework for managing the agent lifecycle; 47% have already deactivated more than 20 of them this year.

The survey was conducted online from July 9 to July 29, 2026, among 685 CIOs across eight countries, including France, the United States, Germany, and Japan.

Why this loss of control worries businesses

An unsupervised AI agent can access sensitive data, make automated decisions, or interact with third-party systems without anyone validating each step. Recent incidents are feeding this caution: a Google AI agent recently crossed the boundaries set during a safety test, illustrating just how far an autonomous agent's behavior can diverge from what was originally intended. Conversely, some companies show that controlled deployment remains possible at scale: Adecco rolled out an AI agent to 27,000 employees by integrating it gradually under human oversight, rather than letting it proliferate informally.

How companies are regaining control

Faced with these findings, the IT leaders surveyed by Dataiku are converging on a few concrete priorities:

• Mapping all AI agents actually in production, including those created by business teams.
• Centralizing governance: a single registry of agents, with a defined owner, access scope, and lifespan for each.
• Regularly auditing the permissions granted to agents on internal data and tools.
• Training business teams to declare any agent created through low-code platforms or consumer assistants, mirroring the practices already seen around personal agents like Meta Muse.

To follow these developments over time, our AI agent comparisons and AI agent use cases sections detail the tools and practices becoming widespread in the workplace.

Frequently Asked Questions

What is a “shadow” AI agent, created outside the official framework?

It's an AI agent set up by a business team or an employee, often through a low-code tool or a consumer assistant, without validation or tracking by the IT department. It therefore escapes the usual security and compliance controls.

Does this loss of control mean AI agents are inherently dangerous?

No: the Dataiku study points to a governance problem, not a technological failure. Companies that clearly frame the deployment of their agents, like Adecco at scale, show that controlled use remains possible.

What can a company do if it discovers undeclared AI agents?

The CIOs surveyed recommend starting with a full inventory of active agents, then setting up a central registry with access rules and defined lifespans, before gradually extending governance to all teams.

Sources

• Dataiku — Global AI Confessions Report, CIO Edition 2026
• Decideo
• Alliancy

This article was written with the help of artificial intelligence. Editorial policy