"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.
Published finding: rapid agent rollout outpaces oversight
Published by Cisco on October 20, the 2026 Splunk State of Observability warned that businesses are deploying AI agents across their networks far faster than they are implementing monitoring procedures and security controls. According to the 2050 ITOps and engineering respondents, 38% said their organization does not yet have clearly defined ownership and accountability for monitoring AI. That gap—measured directly in the report’s survey—places many organizations in a position where new capabilities arrive before governance and visibility do.
Responsibility spans software engineering, infrastructure, web development and information security
The report ties much of the confusion to how responsibility for deployment and monitoring cuts across traditional IT boundaries: software engineering, infrastructure management, web development and information security. This fragmentation makes it difficult to monitor AI use within organizations, and 42% of those surveyed said their main challenge in observing AI is that a different team is responsible for it.
This site is the portfolio.
OSINTSights runs on Cloudflare Workers, D1, R2, and Vectorize, with an AI pipeline on Hetzner ARM. Nubivance designed, built, and operates it. We do the same for clients.
See what we buildHallucinations are widely acknowledged but often not detectable
Respondents were blunt about the practical limits of current observability: while 85% said having visibility into AI hallucinations is important, only 58% said their organization can effectively monitor for potential AI hallucinations. The report notes that for many organizations, identifying when an AI agent is making or acting upon an error is difficult, and they might not have a means of preventing the action from happening—an operational blind spot that can leave real-world processes exposed.
Security consequences: error, exposure and attacker exploitation
The report links monitoring gaps to tangible risks. It explains that a malicious attacker could take advantage of an AI agent to trick it into exposing sensitive information, and that appropriate monitoring tools could prevent that kind of activity. The authors frame this as a convergence problem: in agentic AI, observability and security questions—"Have our agents been compromised? Are they doing their job well?"—are inseparable and happen "at machine speed," as Jeetu Patel, president and CPO at Cisco, put it.
Benefits and trade-offs reported by IT and security teams
Despite the complications, the surveyed organizations report clear gains when agents are deployed correctly. Of those surveyed, 86% said that the deployment of AI had improved the speed at which they are able to detect IT or security issues, and 36% classed that improvement as "significant." Still, the shift is not uniformly positive: 14% suggested that deployment of AI agents had made issues harder to detect and resolve, citing the extra time required to double-check whether an agent's output was correct.
What this means for technologists and security teams, procurement and platform owners
- Technologists and security teams: The report’s data imply an operational imperative to define ownership and enforce cross-team monitoring. With 42% naming cross-team responsibility as a main challenge and only 58% able to detect hallucinations, these teams will have to prioritize tooling and processes that tie observability to controls.
- Procurement and IT leadership: The findings give weight to the call for "common data platforms" that can detect and respond to issues in real time. As Jeetu Patel said, "Common data platforms that help us detect and respond to issues in real-time are the key to deploying agents we can trust at scale."
- Vendors and platform owners: The report underscores customer demand for observability and security to converge. Vendors will be measured against their ability to surface hallucinations, provide cross-team visibility, and integrate controls that prevent agents from acting on errors.
Splunk’s report sketches a simple operational dilemma: agentic AI can speed detection and automate routine tasks, but those benefits arrive alongside new failure modes that current organizational structures and tooling often cannot contain. As the report highlights, the technical solution space—common data platforms, integrated observability and security—exists as a prescription; the harder question is whether organizations will close the accountability gaps the survey exposes before an avoidable incident forces the issue.
