Agent Inventory Blind Spot: 18% of Enterprises Know What's Running | Ariana.Digital · Enterprise Agentic AI Insights
IBM says enterprises will run 1,600+ AI agents by year-end. Only 18% have a complete inventory. ServiceNow's AI Control Tower just added a kill switch — but it only works if you know what to watch. Data brief + 3-move action checklist.
The ServiceNow AI Control Tower just got a kill switch. That headline got a lot of play at Knowledge 2026. But here is the thing most of the coverage missed: the kill switch only works if you know which agents to watch. IBM's latest enterprise survey says large organizations will run over 1,600 AI agents by the end of this year. The OutSystems 2026 State of AI Development report, which surveyed 1,900 IT leaders, found that only 18% of those organizations maintain a complete, current inventory of the agents already running inside their walls. Only 12% have a centralized platform to manage the sprawl. So the kill switch problem is actually an inventory problem first. Here is what that looks like in practice for a CIO running a mid-to-large enterprise right now. You have Now Assist handling ITSM ticket deflection inside ServiceNow. You have Microsoft 365 Copilot and Copilot Studio agents running across productivity workflows. You have two or three Bedrock-backed agents your platform team spun up for internal tooling. You have at least one Agentforce pilot running in Service Cloud. None of these have a shared audit trail. None of them are visible from a single pane. If one of them starts making decisions it should not be making, your first signal is probably a ticket, not a dashboard. ServiceNow's move at Knowledge 2026 was to change that. The expanded AI Control Tower now covers 30 enterprise integrations including AWS, Google Cloud, Azure, SAP, Oracle, and Workday. It can discover, observe, govern, secure, and measure AI deployed across any of those systems, not just inside the ServiceNow estate. The enforcement piece — the actual kill switch — enters GA in August. August matters for another reason. August 2 is when the EU AI Act GPAI enforcement provisions go live. The European Commission can start applying fines against GPAI model providers and the enterprises deploying them without governance documentation. That deadline did not move in the Omnibus agreement reached on May 7. The high-risk Annex III deadline did move (pushed from August 2026 to December 2027), and that has created some board-level "we have more time" energy that is worth pushing back on directly. The GPAI obligations — agent action logging, data access documentation, and training data transparency — are on their original schedule. The CRO sitting next to you in the next QBR is going to have a question: can you show an auditor which agents accessed which data, when, and what decisions they influenced? If the answer is "we'd have to pull that manually from four different systems," you are in a position that is harder to defend than it was 12 months ago. The data picture right now: | Stat | Source | |------|--------| | 1,600+ agents per large enterprise by year-end | IBM Think 2026 | | 18% have a complete agent inventory | OutSystems 2026 (1,900 IT leaders) | | 12% have a centralized management platform | OutSystems 2026 | | 94% say sprawl is increasing risk | OutSystems 2026 | | 70% say current governance is not fit for purpose | IBM Think 2026 | | ServiceNow AI Control Tower: 30 integrations, GA August 2026 | ServiceNow Knowledge 2026 | | EU AI Act GPAI enforcement: August 2, 2026 | EU AI Act Chapter V | The move Three things a CIO or CDO can do this week without waiting for the August GA: 1. Run the inventory. Pull a list of every AI/agent license, integration, or custom deployment across your major platforms. This is not a governance project. It is a spreadsheet with five columns: platform, agent name, data it can access, actions it can take, who owns it. If you cannot fill it in under two hours, that tells you something. 2. Map the blast radius. For each agent, ask: what is the worst case if this goes off-script? Agents with access to customer data, financial systems, or employee records should be first in line for a governance layer. 3. Get eyes on the ServiceNow AI Control Tower August timeline. If you are already running Now Assist or ServiceNow workflows, the cross-platform governance layer is coming to your environment. Get your vendor team to walk you through the August GA scope now so you are not scrambling in September. The board question about AI ROI and AI risk is the same question. The organizations that have the inventory have an answer to both. --- Sources: ServiceNow Knowledge 2026 · IBM Think 2026 — 1,600 agents per enterprise · OutSystems 2026 State of AI Dev · EU AI Act Omnibus, EU Council May 7 2026 · Fortune: ServiceNow kill switch · Latham & Watkins: Omnibus timeline