News
AI Agents Are Moving Faster Than Enterprise Security Controls, Study Finds
Enterprises are deploying AI agents that can act within critical business applications, but security and governance controls are struggling to keep pace, according to new research from Delinea.
The cybersecurity company’s 2026 Application Access Governance Report, published October 27, surveyed 1,000 senior finance and finance technology decision-makers in the United States and United Kingdom.
The report found that 35% of surveyed organizations had deployed AI agents capable of taking actions within critical business applications, including enterprise resource planning (ERP) and financial systems.
Among the 879 respondents whose organizations use AI tools or agents in critical applications, 44% said deployment was advancing faster than governance and security controls. Only 11% said those controls were keeping pace with AI adoption.
The findings highlight the oversight challenges associated with agents that can execute transactions and modify business records, particularly when organizations cannot trace their actions or consistently identify who is responsible for them.
AI Agents Introduce New Operational Risks
Eighty-four percent of respondents using AI in critical business applications reported at least one AI-related issue during the previous 12 months. The problems extended beyond inaccurate AI-generated information.
Thirty percent reported incorrect financial transactions linked to AI, while 28% reported unintended or unauthorized actions. Another 32% said they could not trace or explain AI actions.
The research also found that 28% had experienced situations in which AI agents continued operating after the employees who deployed them had left the organization.
Agents capable of acting across connected applications require oversight beyond reviewing generated information. Organizations also need to manage their permissions and monitor what they do.
Enterprise Governance Controls Remain Inconsistent
The report also identified weaknesses in existing application access governance practices. Only 46% of surveyed organizations continuously monitored access risks across their critical business applications, while 22% maintained a single named owner for each application in a centralized system.
Despite those gaps, 95% of respondents were confident their access controls could withstand external audit scrutiny. However, 26% reported recurring audit findings, and 25% said some findings had remained unresolved for three years or longer.
Delinea said organizations need to extend established identity and access governance practices to AI agents. These include least-privilege permissions and regular access reviews, along with segregation-of-duties controls that separate responsibilities that should not be combined.
Frank Vukovits, principal evangelist at Delinea, said organizations remain responsible for actions taken by AI agents within their systems.
“When an agent can approve a transaction or change permissions, the business owns the outcome. You need to know what every agent can do, who owns it and who approved it, before an auditor asks or your financials are misstated,” Vukovits said.