Companies are introducing AI agents into their networks far more quickly than they can put monitoring systems and security safeguards in place, according to a report published by Cisco on October 20. The 2026 Splunk State of Observability found that while many firms are eager to tap into the potential advantages of using AI in their operations, a large number are doing so without a plan to make sure it's being used securely. The result is exposure to security risks, hallucinations, and other typical problems tied to the technology.
Among the 2,050 ITOps and engineering professionals who responded, 38% reported that their company does not yet have well-defined ownership and responsibility for overseeing AI. Much of this lack of clarity comes from the way responsibility for launching and tracking AI agents spans traditional divisions within IT departments, including software engineering, infrastructure management, web development, and information security. This splintering has created obstacles to tracking AI use inside the organization, and 42% of survey participants said their biggest hurdle in observing AI is that a separate team handles it. Meanwhile, even though 85% of respondents indicated that having insight into AI hallucinations matters, only 58% said their company can successfully monitor for possible AI hallucinations. Many organizations find it hard to spot when an AI agent is generating or acting on an error, and they may lack tools to stop the action from taking place.
"When agent observability is everyone's job, it becomes nobody's job," said Mark Woods, field CTO and chief technical advisor for EMEA at Splunk. The report notes that managing agents at scale demands rethinking both security and observability, as questions about whether agents have been compromised and whether they're performing well are now intertwined and moving at machine speed. Jeetu Patel, president and CPO at Cisco, stated that observability and security data are converging, and teams need to work together more than ever before. Common data platforms that help detect and respond to problems in real-time are essential for deploying agents that can be trusted at scale, he added. The report also found that 86% said AI deployment had improved the speed at which they can spot IT or security issues, with 36% describing that improvement as significant. However, 14% suggested that rolling out AI agents had made issues tougher to detect and fix.
The confusion over who owns AI monitoring stems from the technology's ability to cut across long-established IT silos, according to the report. When no single team clearly controls deployment and oversight, visibility gaps emerge that attackers can exploit—for instance, by manipulating an AI agent to leak sensitive data. Without the right monitoring infrastructure, such activity could go unnoticed until damage is done. The report warns that organizations rushing to adopt AI without corresponding observability tools are operating blind, unable to distinguish between helpful automation and harmful errors. One of the main concerns cited was the additional time required to verify whether what the agent was producing was accurate, undermining the efficiency gains AI is supposed to deliver.
When AI agents are implemented properly, they can offer a meaningful boost to IT and security teams, the report concludes. But the current gap between adoption speed and monitoring capability leaves many organizations vulnerable to threats they can't yet see or control. The report recommends that firms establish clear accountability for AI oversight, invest in unified data platforms that bridge security and observability, and build processes to validate AI outputs before acting on them. The bottom line: speed without visibility is risk, and companies need to match their enthusiasm for AI with an equal commitment to tracking how it behaves. Organizations that fail to close this gap may find that the efficiency they gain from AI agents is offset by the security incidents and operational errors they can't prevent. The shift toward agentic AI requires a cultural and structural change within IT departments, not just new tools.