Kore.ai survey says firms face unmanaged AI agent risk

Kore.ai has published a survey showing that most large enterprises believe their AI agents operate with unmanaged risk. The study covered more than 400 IT business leaders at US organisations with 2,000 or more employees.

The findings highlight a gap between the pace of AI agent deployment and the controls companies have in place to oversee those systems. According to the survey, 72% of respondents said their AI agents create unmanaged financial or compliance risk, while 79% said they had to reverse an action taken by an AI agent.

Another 70% said they had experienced a failure their teams could not trace, and 62% said governance concerns had delayed deployments. More than half, 53%, said they were running agents they did not fully trust or understand, while 42% reported lost revenue linked to an AI agent failure.

The report suggests the operational impact can spread beyond a single system. Four in 10 respondents said one agent failure had cascaded across multiple systems, extending the effects of a bad decision into other parts of the business.

Where Agents Are Used

The survey also sheds light on the work companies are assigning to AI agents. It found that 41% of agents are used for data migrations and system updates, 26% for approving or denying decisions, and 15% for financial transactions.

These figures suggest AI agents are not limited to low-stakes administrative tasks. Many are being placed in roles that affect core operations, customer interactions and financial processes, increasing the consequences when controls fail or oversight is weak.

Kore.ai said this pattern reflects a broader shift in enterprise AI from experimentation to direct operational use. The report argued that many businesses are giving AI systems authority over decisions and processes without clear visibility into how those systems act once deployed.

Governance Question

Kore.ai drew a distinction between monitoring AI agents after deployment and building governance into them from the outset. Its position is that a policy layer or monitoring tool added later can watch a running agent, but cannot shape how the agent was originally designed or how later updates are introduced.

This view comes as businesses face growing pressure to show that AI systems can be audited, traced and controlled. The survey results suggest many technology leaders do not yet believe they have met that standard, despite moving ahead with deployment.

Outside research cited in the report points in the same direction. McKinsey has reported that 62% of organisations are experimenting with or scaling AI agents, but fewer than 10% have scaled them in any single function. Gartner has forecast that by 2027, 40% of enterprises will demote or decommission autonomous AI agents because of governance failures.

Platform Push

Kore.ai used the findings to frame its approach to agent management. Its Artemis edition of the Kore.ai Agent Platform is designed to cover building, deployment, management and optimisation in one system, with governance checks applied before an agent goes live.

The platform includes an AI agent architect called Arch, which turns plain-language objectives into agents defined in Agent Blueprint Language. According to Kore.ai, that language validates and governs each agent before deployment, while the same system manages traceability, policy enforcement and later improvement in production.

Kore.ai also said the platform is model-agnostic and cloud-agnostic, and can run in public cloud, sovereign, private or on-premises environments with regional data residency. It added that Arch reviews production behaviour and proposes improvements for human approval.

The survey was conducted by Propeller Insights on behalf of Kore.ai. Respondents were IT business leaders aged 34 to 50 at large US organisations, and the stated margin of sampling error was plus or minus 3.0 percentage points at a 95% confidence level.

Raj Koneru, Chief Executive Officer and Founder of Kore.ai, linked the findings to a wider change in how companies judge enterprise AI. "Enterprise AI has shifted from showing that AI works to proving it can be trusted," said Raj Koneru, Chief Executive Officer and Founder of Kore.ai. "Governance has to be built into the agent itself, not added once it is running, because trust comes from visibility, reproducibility, auditability, and control, not from the model getting it right every time. The companies that scale AI will be the ones using AI to build, govern, and improve AI on a single layer. That is the architecture this market is moving to, and it is the one we've built."

Peter Mullen, Chief Marketing Officer at Kore.ai, made a similar point about the limits of oversight tools alone. "The market is solving for visibility, but enterprises need accountability. Those are not the same thing. An agent that can be watched but not governed is still a liability. Enterprises do not need better ways to monitor their agents. They need agents built right from the start and governed through production scale," Mullen said.