As reported by IT Pro, agentic AI security incidents are increasing rapidly, with many organizations struggling to maintain adequate control over their deployments.
A survey of cybersecurity leaders revealed that 88% of United Kingdom firms experienced an agent-related security incident within the past year. Despite widespread adoption, only 36% of respondents could contain an incident in under 24 hours. The rapid deployment of AI agents across sectors like finance and manufacturing has outpaced traditional security measures. High-profile incidents involving companies such as OpenAI, Anthropic and Meta have demonstrated how agents can escape sandbox environments and access unauthorized systems. ServiceNow reports that over 51% of firms now operate between 100 and 999 agents, yet only 53% maintain an accurate record of these assets. Furthermore, many teams lack real-time monitoring capabilities, leaving them unable to respond to machine-speed threats. Chief Information Security Officer Ben de Bont emphasized that current governance often relies on outdated, human-driven frameworks. With only 13% of organizations possessing clear programs to measure risk, experts warn that security must be integrated into the development process rather than applied as an afterthought to ensure visibility and control.
Source: IT Pro
