81% of CIOs Lack Full Oversight of AI Agents Built Outside IT

How Shadow AI Development Is Reshaping IT Governance

Most chief information officers cannot fully monitor the AI agents their employees create outside approved systems, according to a survey conducted by Dataiku. The Harris Poll carried out the online survey between July 9 and July 29, gathering responses from 685 CIOs across the United States, the United Kingdom, and France. The findings reveal a widespread gap in visibility over AI tools developed beyond formal IT channels.

The survey shows that 81% of participating CIOs admitted they lack complete oversight of AI agents built by staff outside sanctioned processes. This highlights a growing challenge as departments increasingly adopt AI tools independently, often without central governance. Dataiku noted that while such grassroots innovation can drive agility, it also introduces risks related to security, compliance, and integration when IT teams cannot track or manage these systems.

What Steps Can Organizations Take to Regain Control?

The rise of decentralized AI creation reflects broader trends in workplace technology adoption, where employees bypass traditional approval routes to solve problems quickly. Many CIOs expressed concern that this trend complicates efforts to maintain consistent data standards and protect sensitive information. Without visibility into how these agents are built or what data they access, organizations face potential vulnerabilities in auditing and regulatory compliance. The survey suggests that current governance models struggle to keep pace with the speed at which teams experiment with AI tools.

Experts recommend establishing clear guidelines for AI use that balance innovation with oversight, rather than imposing outright bans that may push activity further underground. Some companies are adopting AI registries or lightweight approval workflows to track tools without stifling experimentation. Dataiku emphasized the importance of collaboration between IT and business units to create shared responsibility for AI safety and performance. The goal, they said, is to enable innovation while ensuring accountability across the organization.

What percentage of CIOs reported lacking full oversight of externally built AI agents? Eighty-one percent of the 685 CIOs surveyed said they do not have complete visibility into AI agents created by staff outside formal IT channels.

Frequently Asked Questions

Which countries were included in the Dataiku survey on AI oversight? The survey included chief information officers from the United States, the United Kingdom, and France, conducted online by The Harris Poll between July 9 and July 29.

Why is oversight of independently developed AI agents a concern for IT leaders? Without visibility, IT teams cannot assess security risks, ensure data compliance, or integrate these tools effectively, potentially exposing organizations to operational and regulatory challenges.