New research from Kore.ai found that 72% of enterprise leaders believe the AI agents operating in their organisations introduce unmanaged financial or compliance risk. The report, which includes a litany of other such concerning statistics, adds to a growing body of evidence suggesting that agent adoption is accelerating faster than the controls needed to manage it. CallMiner recently warned that AI adoption is outpacing governance maturity across European customer experience teams, while Sinch also reported that nearly three quarters of organisations had rolled back a customer-facing AI agent due to governance concerns.
Enterprises Are Deploying Agents They Do Not Fully Trust
The Kore.ai survey paints a picture of organisations moving forward with agentic AI despite significant uncertainty about how those agents behave in production. Eighty-two per cent of respondents reported AI agents have autonomously executed consequential actions in production, and 79% of those actions required manual reversal. Ninety-three per cent described the reversal process as costly and disruptive, and 53% admitted deploying an agent without fully understanding or trusting how it would behave. The challenge is no longer proving that AI agents can perform useful work, but ensuring they can do so safely, consistently and predictably at scale.
Visibility Remains a Major Weakness
Perhaps the most concerning finding relates to observability. While most organisations can detect that something has gone wrong, 70% revealed they could not identify which agent was responsible when a failure occurred in a multi-agent environment. This lack of traceability complicates everything from root cause analysis to regulatory reporting, leaving compliance teams unable to demonstrate accountability when an agent's decision is challenged. As enterprises move from isolated copilots to interconnected agent ecosystems, knowing why a decision was made and which agent made it is increasingly becoming as important as the decision itself.
Why This Matters for Customer Experience
For CX leaders, agent failures are rarely confined to IT. The survey uncovered that 42% of organisations reported revenue loss following an AI agent failure, 28% experienced customer churn, and 31% reported SLA violations. Customer trust, once eroded by a visible AI failure, is difficult to rebuild, and the reputational fallout often outlasts the technical fix itself. These outcomes point to the emerging reality that AI governance has moved beyond a technology issue to becoming a customer experience, revenue and brand trust issue.
Governance May Become the Next AI Battleground
The survey also helps explain a broader market shift. Vendors are increasingly positioning governance, monitoring and agent lifecycle management as strategic differentiators, including Kore.ai's own recently launched Artemis platform. Traditional governance models, built for deterministic software, are unlikely to be sufficient for autonomous agents operating across multiple systems, according to Gartner. The analyst firm recently warned that one-size governance could undermine enterprise AI agent programmes.
The question facing enterprises is changing. It is no longer how quickly organisations can deploy AI agents, but whether they can govern hundreds, or eventually thousands, of autonomous agents once they are running in production. If recent research from Kore.ai, Sinch, CallMiner and Gartner points to anything, it is that governance is emerging as a defining challenge of CX AI and enterprise AI more broadly.