AI Agents Are Racing Ahead of the Oversight Meant to Keep Them Safe

Businesses are rolling out AI agents across their networks much faster than they can put monitoring and security controls in place. The 2026 Splunk State of Observability report, published by Cisco on October 20, warns that many organisations are eager to gain the benefits of AI but lack a strategy to ensure it is used safely. This leaves them exposed to security threats, hallucinations (when AI produces false or misleading output) and other common problems.

A key issue is accountability. Of the 2050 IT operations and engineering respondents, 38% said their organisation has no clearly defined ownership for monitoring AI. Responsibility for AI agents cuts across software engineering, infrastructure, web development and information security, and 42% said their main challenge is that a different team is responsible for observing AI. As Splunk's Mark Woods put it: "When agent observability is everyone's job, it becomes nobody's job."

There is also a gap in visibility. While 85% of respondents said insight into AI hallucinations is important, only 58% said their organisation can effectively monitor for them. Spotting when an agent acts on an error is hard, and some organisations may have no way to stop the action. The report notes that an attacker could trick an AI agent into exposing sensitive information, and that appropriate monitoring tools could help prevent this.