ServiceNow AI Control Tower, Action Fabric, and Autonomous Workforce: The Complete Guide for Agentic AI Architects (2026)

When ServiceNow’s Nenshad Bardoliwalla drew the line at Knowledge 2026 in Las Vegas — “Others let agents read and write data. We let agents execute governed work” — he was naming the problem every enterprise automation team already felt: the gap between deploying AI agents and trusting them to run real business processes unsupervised.

That gap is now the most contested territory in enterprise software. Salesforce launched Headless 360 at TDX 2026. Microsoft extended Copilot into always-on Autopilot agents. Google shipped its Gemini Enterprise Agent Platform. But ServiceNow’s Knowledge 2026 announcements — AI Control Tower expanded into a five-function governance layer, Action Fabric exposing the platform’s system of action to any AI agent via Model Context Protocol (MCP), and an Autonomous Workforce of domain-scoped AI specialists — represent the most architecturally complete attempt yet to answer the question: who governs the agents after they’re deployed?

This guide breaks down every major announcement, the technical architecture behind each, what’s generally available now versus roadmapped, and what it means for RPA teams, solution architects, and agentic AI practitioners planning their next move.

Why ServiceNow’s Bet Matters: From System of Record to System of Action

ServiceNow processes over 100 billion workflows and 7 trillion workflow transactions annually across its customer base. That transaction volume isn’t just a vanity metric — it’s the foundation of the company’s argument that it already owns the enterprise’s operational graph: the relationships between people, processes, assets, policies, and decisions that determine how work actually gets done.

The strategic pivot announced at Knowledge 2026 reframes ServiceNow from a workflow automation platform into what the company calls a “platform of platforms” — an end-to-end operating system where AI agents from any vendor can discover, request, and execute governed enterprise actions. The shift is subtle but significant: analyst Derek du Preez of diginomica described it as moving from “automated workflows” to “automated work,” a distinction that separates orchestrating predefined steps from agents that autonomously complete business outcomes.

For RPA practitioners, this matters directly. The Forrester Wave for Adaptive Process Orchestration Q3 2026 now evaluates vendors on exactly this capability — whether their platforms can govern autonomous agents alongside traditional process automation. ServiceNow’s answer is a four-pillar AI Platform: Sense (data fabric), Decide (context engine), Act (autonomous workforce), and Secure (AI Control Tower).

AI Control Tower: The Five-Function Governance Layer

AI Control Tower was first introduced at Knowledge 2025 as a visibility and management tool. At Knowledge 2026, ServiceNow expanded it into an end-to-end governance offering organized around five functions. Each addresses a specific failure mode that enterprises hit as agent deployments scale beyond pilot programs.

Discover: Finding Every AI Asset in the Enterprise

Most enterprises cannot answer a basic question: how many AI models, agents, and automations are running across the organization right now? AI Control Tower’s Discover function attacks this blind spot with 30 new integrations spanning AWS, Google Cloud, Microsoft Azure, SAP, Oracle, and Workday.

More Read

AMD Helion: The Open Rack Blueprint for Agentic AI (2026)

UiPath vs Automation Anywhere vs SS&C Blue Prism: The 2026 Agentic Platform Showdown

RPA Security: Safeguarding Your Digital Workforce

What makes this more than a simple asset inventory is its scope. Discovery covers not just AI models and agents but also non-human identities — service accounts, bots, API keys — and connected OT and IoT devices. The foundation is ServiceNow’s Configuration Management Database (CMDB), which maps digital assets to the services, people, and processes they support. For organizations running both legacy RPA bots and new agentic AI deployments, this creates a single registry that doesn’t treat attended bots, unattended bots, and autonomous agents as separate governance categories.

Observe: Runtime Monitoring That Replaces Periodic Audits

ServiceNow’s acquisition of Traceloop — an observability company specializing in LLM and agent tracing — powers the Observe function. Instead of reviewing agent behavior through periodic manual audits (the current practice at most enterprises), Observe provides continuous runtime monitoring with live metrics and alerts.

The practical value is visibility into how agents reason and make decisions: which inputs they considered, which steps they took, and where failure points occurred. For agentic AI architects building multi-agent systems, this is the difference between knowing an agent failed and understanding why it failed — which chain-of-thought step broke, which tool call returned unexpected results, which policy guardrail was triggered or missed.

Govern: Risk Frameworks Aligned to NIST and the EU AI Act

The Govern function applies AI-driven risk assessment to agents, models, datasets, prompts, and classic machine learning models. Five new risk frameworks aligned with the NIST AI Risk Management Framework and the EU AI Act provide compliance controls that enterprises can configure to their regulatory environment.

This is where ServiceNow’s enterprise workflow heritage becomes an architectural advantage. Governance isn’t a separate reporting layer — it’s embedded in the same workflow engine that runs the agents. Automated risk assessment triggers remediation workflows, not just alert emails. For RPA teams familiar with governance as a spreadsheet exercise, the shift to embedded, automated governance mirrors the broader shift from attended RPA to autonomous agentic work.

Secure: The Kill Switch for Rogue Agents

The Knowledge 2026 keynote’s most memorable demo was a live demonstration of what ServiceNow calls real-time agent containment. Using Veza’s access graph (ServiceNow acquired Veza and closed the deal in March 2026), AI Control Tower detected an agent compromised by prompt injection, mapped its blast radius across connected systems, revoked its permissions, shut it down, and generated a P1 security incident with full audit documentation — all in real time.

The Secure function extends identity access governance to hyperscaler AI environments and connected devices using scoped permissions and least-privilege enforcement. Veza’s technology maps over 30 billion fine-grained permissions across human, non-human, and agent identities. When a vendor updates an AI model or an agent’s tool configuration changes, the system automatically re-scopes permissions rather than waiting for a human review cycle.

For practitioners building systems where agent safety and governance are non-negotiable — financial services, healthcare, government — the Secure function addresses the specific threat model that keeps CISOs up at night: an autonomous agent with valid credentials operating outside its intended scope, either through compromise or drift.

Measure: Cost Tracking Before Runaway Spend Becomes a Crisis

The Measure function provides cost tracking and ROI dashboards that link AI usage (by token, by workflow, by agent) to business outcomes and spending. ServiceNow reports tracking over 1,600 internal AI assets and measuring half a billion dollars in cumulative AI value from its own deployments in 2025.

This addresses the “shadow AI” cost problem that parallels shadow IT a decade ago. As teams spin up agents using different LLM providers, token costs accumulate across budgets that weren’t designed for consumption-based AI pricing. Measure gives finance and IT a single view before the quarterly surprise.

Action Fabric: Opening the System of Action to Every AI Agent via MCP

If AI Control Tower is the governance layer, Action Fabric is the execution layer. It exposes ServiceNow’s workflows, approval chains, playbooks, catalog requests, and business rules to any AI agent — whether built on ServiceNow, Anthropic’s Claude, Microsoft Copilot, or a customer’s own agent framework — through a generally available Model Context Protocol (MCP) Server.

The distinction ServiceNow draws is between data access and governed execution. Many enterprise platforms let agents read and write data through APIs. Action Fabric lets agents execute governed work: trigger a password-reset playbook from Claude, start an onboarding workflow from a collaboration tool, or process an invoice dispute from a CRM agent — all running through the same approval chains, business rules, and audit trails that human-initiated work follows.

How the MCP Server Works

The Model Context Protocol, which joined the Agentic AI Foundation alongside Google’s A2A protocol, is the open standard ServiceNow chose for agent-to-platform communication. The ServiceNow MCP Server is included in every Now Assist and AI Native SKU — no additional licensing required.

Actions run “headlessly,” meaning directly in the platform without requiring the ServiceNow user interface. The technical flow works like this:

1. Agent connects via the MCP Server, which manages OAuth authentication and session management.
2. Context Engine supplies situational awareness — the Knowledge Graph maps relationships between the requester, affected systems, relevant policies, and historical decisions.
3. Action executes through governed workflows — playbooks, business rules, and approval chains run exactly as they would for a human-initiated request.
4. AI Control Tower audits the entire transaction — identity verification, permission scoping, and full audit trail.
5. Operational data feeds back into process mining, the CMDB, and analytics, improving context for future actions.

MCP Server Console: Governance at the Protocol Level

The MCP Server Console provides enterprise-grade controls that go beyond basic API gateway management:

The MCP Server spans IT, HR, customer service, security, risk and compliance, and app development — supporting the coordinated multi-agent systems that Microsoft’s Agent Framework 1.0 and other orchestration layers are designed to coordinate.

Anthropic as First Design Partner

Anthropic is the first named design partner for Action Fabric, connecting Claude Cowork directly to ServiceNow’s system of action. In the reference scenario demonstrated at Knowledge 2026, a new product manager using Claude Cowork after a role change has access gaps identified automatically — missing repository access, Slack channel memberships, and tool provisioning — with each request routed through the appropriate ServiceNow approval chain.

Boris Cherny, Head of Claude Code at Anthropic, framed the partnership this way: “The gap between knowing what needs to happen and making it happen is where productivity dies.” Action Fabric closes that gap by giving Claude (and any MCP-compatible agent) the ability to not just identify what needs to happen but execute it through governed enterprise workflows.

Autonomous Workforce: AI Specialists That Complete End-to-End Processes

ServiceNow’s Autonomous Workforce moves beyond the “copilot” paradigm — where AI assists humans with individual tasks — to “AI specialists” that own and complete entire business processes. Each specialist has a designated domain, a human manager, performance metrics, and accountability for outcomes.

Currently Available AI Specialists

L1 IT Service Desk Specialist — generally available. Handles routine IT requests (password resets, access provisioning, software installations) end-to-end without human escalation. Standard agents handle isolated tasks; the specialist chains multiple agents to resolve complete workflows.

CRM Specialists — generally available across sales and service. In the Sales CRM suite, a specialist manages a deal from customer inquiry to signed contract, including CPQ pricing, quote generation, internal approvals, and order fulfillment. In the Service CRM suite, Autonomous Service handles customer requests across voice, chat, and digital channels. In one demo scenario, one specialist takes a modem outage report, another schedules a repair, and a third checks parts availability for the field technician.

Roadmap: IT, Security & Risk, and Beyond

IT Specialists — expected in June 2026, covering IT operations, AIOps, SRE, and asset lifecycle management.

Security and Risk Specialists — entering preview in June 2026, with general availability targeted for September 2026. These are the first full integration of the Armis and Veza acquisitions, with two new AI specialists handling vulnerability resolution and security operations end-to-end. ServiceNow’s own security team reports handling incidents seven times faster than prior workflows using these specialists internally.

Additional specialist areas announced include HR, finance, legal, procurement, workplace services, supplier management, portfolio planning, and health and safety. Timelines for these were not specified beyond “through 2026.”

Architecture: How Specialists Differ from Standard Agents

All specialists share three platform components:

• Workflow Data Fabric — connects to data sources across the enterprise, including new partnerships with Experian and Boomi, providing real-time operational context.
• CMDB and Context Engine — maps digital assets to services, people, and processes, so specialists understand not just what they’re doing but why it matters and what it affects.
• AI Control Tower — governs every specialist action with the same five-function oversight applied to any other agent on the platform.

The structural difference from standard agents is scope. A standard agent handles an isolated task — resetting a password. A specialist owns a process outcome — resolving an employee’s inability to access a system, which might involve checking permissions, resetting credentials, verifying MFA enrollment, updating the access request record, and notifying the employee’s manager, all without human intervention.

Otto: The Conversational Front Door to the Enterprise

Otto combines three previously separate ServiceNow experiences — EmployeeWorks (based on the $2.85 billion Moveworks acquisition), Now Assist, and AI Experience — into a single multimodal interface. It supports four interaction modes:

1. Conversational AI — natural language requests for IT, HR, procurement, and other enterprise services.
2. Enterprise Search — unified search across ServiceNow and connected systems, including SharePoint.
3. Multilingual AI Voice Agents — voice-based interaction for contact center and employee service scenarios.
4. AI Data Explorer — plain-language analytics queries against enterprise data.

Otto is available now in EmployeeWorks and AI Control Tower, with rollout across all ServiceNow products planned through the remainder of 2026. For organizations that have struggled with portal adoption — a chronic challenge in ITSM — Otto replaces “navigate to the right catalog item” with “describe what you need in natural language.”

The Acquisition Strategy: $11+ Billion in Capability Bolts

ServiceNow’s Knowledge 2026 announcements are inseparable from the acquisition strategy that funds them. Four acquisitions, totaling over $11 billion, provide the technical foundation for the AI Platform’s governance and security capabilities:

The Armis acquisition is ServiceNow’s largest ever, and it signals where the company sees the governance boundary expanding. Traditional IT governance covers servers, applications, and user identities. The combination of Armis (what exists) and Veza (who can access it) extends that boundary to OT controllers, IoT sensors, medical devices, and AI agents — creating what ServiceNow calls “a single graph that maps every identity, every permission, and every connected asset.”

Partnership Ecosystem: Microsoft, NVIDIA, and Anthropic

Microsoft: Governing Agent 365 from ServiceNow

Building on the November 2025 partnership that linked AI Control Tower with Microsoft Foundry and Copilot, the Knowledge 2026 expansion has two dimensions:

• Governance integration — AI Control Tower extends its discovery, observation, and security functions across the Microsoft Agent 365 ecosystem, so ServiceNow-governed enterprises get a single governance view across both platforms.
• Marketplace presence — ServiceNow AI specialists will appear in the Agent 365 Marketplace as “digital employees,” subject to Microsoft’s permissions and admin controls. Preview is available now, with marketplace availability expected later in 2026.

NVIDIA: Desktop Agents and Benchmarking

The NVIDIA partnership has three components:

• Project Arc — an autonomous desktop agent running on the NVIDIA OpenShell sandboxed runtime, currently in early preview. This combines ServiceNow’s governance and workflow capabilities with NVIDIA’s compute infrastructure for agents that operate a desktop as a human would — a direct overlap with traditional RPA attended-bot territory.
• Enterprise AI Factory — AI Control Tower is included in the NVIDIA Enterprise AI Factory validated design, providing agent observability and governance at the infrastructure layer.
• NOWAI-Bench — an open-source benchmark comprising EnterpriseOps-Gym and EVA-Bench, designed to evaluate agent performance in enterprise operational scenarios.

Anthropic: Claude Cowork as the First MCP Design Partner

As detailed in the Action Fabric section, Anthropic’s Claude Cowork is the first agent framework connected to ServiceNow’s MCP Server. This partnership positions ServiceNow as the governed execution layer that Claude delegates to when enterprise actions are required — an architecturally clean separation between the reasoning layer (Claude) and the action layer (ServiceNow).

Competitive Landscape: Who Else Wants to Be the Agent Control Layer?

ServiceNow is not alone in claiming the enterprise agent governance position. Here’s how the major contenders compare:

The battle lines are clear: every major enterprise platform wants to be the layer that governs AI agents. ServiceNow’s argument is that governance requires operational context — you can’t properly govern an agent’s action without understanding the process it’s participating in, the assets it’s touching, and the identities it’s impersonating. That context lives in the CMDB and the workflow transaction history, which ServiceNow has been building for two decades.

Analysts are cautiously supportive. Charles Betz of Forrester noted that “adding security data to the knowledge graph extends ServiceNow’s CMDB roots naturally” and that “companies still want deterministic, auditable workflows.” Rebecca Wettemann of Valoir highlighted the proactive alerting in the AI Control Tower demo as standing out among enterprise agent orchestration platforms. Thomas Wieberneit, an independent consultant, observed that most customer adoption “remains scenario-based, often starting with service use cases.”

For RPA practitioners evaluating which platform to standardize on, the choice depends on where your organization’s operational center of gravity sits. If your ITSM and workflow backbone is already ServiceNow, Action Fabric and AI Control Tower extend your existing investment into agentic governance. If your automation center of gravity is in UiPath or Automation Anywhere, you’ll want to evaluate how ServiceNow’s governance layer interoperates with your existing bot estate — the 30 new Control Tower integrations suggest ServiceNow is positioning itself as a governance overlay, not a replacement.

What’s Available Now vs. What’s Roadmapped

One of the most important questions for any enterprise announcement is “when can I actually use this?” Here’s the availability breakdown as of October 2026:

The critical takeaway: the core platform — Action Fabric with MCP Server, AI Control Tower, and the initial AI specialists — is generally available now. This isn’t a vision deck. The consumption model uses the same Assist currency used for Now Assist and AI Agents, which means organizations already using Now Assist don’t need a separate procurement cycle to start using Action Fabric.

What This Means for RPA Teams and Automation Architects

ServiceNow’s Knowledge 2026 announcements have five direct implications for teams currently running RPA programs or building agentic AI capabilities:

1. The governance gap is now a product category. If your organization runs AI agents from multiple vendors — and increasingly, most do — you need a governance layer that works across all of them. AI Control Tower’s cross-platform discovery and monitoring addresses the same “bot sprawl” problem that plagued early RPA programs, but at the agent level. The pattern is familiar: ungoverned automation scales until it breaks, then the cleanup costs more than doing it right from the start.

2. MCP is becoming the enterprise integration standard for agents. ServiceNow joining Anthropic, Salesforce, and dozens of other enterprise vendors in supporting MCP as the agent-to-platform protocol reinforces what the AAIF standardization effort set in motion. For architects designing multi-agent systems, MCP competency is no longer optional — it’s the integration layer your agents will use to interact with enterprise platforms.

3. The “attended bot” is becoming the “desktop agent.” NVIDIA’s Project Arc — autonomous desktop agents running on OpenShell with ServiceNow governance — directly overlaps with traditional attended RPA territory. The difference is autonomy: attended bots follow scripts, desktop agents reason about tasks. RPA teams should expect this category to mature rapidly as NVIDIA’s compute infrastructure and ServiceNow’s governance layer provide the production-grade foundation that earlier desktop agent experiments lacked.

4. Headless execution changes the integration architecture. Action Fabric’s headless workflow execution means AI agents don’t need to navigate ServiceNow’s UI — they can trigger governed actions directly through the MCP Server. For RPA teams that currently use screen-scraping or UI-based integrations with ServiceNow, this offers a more reliable, API-native integration path. It also means your existing ServiceNow workflows become agent-accessible without rewriting them.

5. Governance-first is becoming the winning enterprise sales motion. ServiceNow’s strategy — win on governance and trust first, expand the platform argument second — mirrors what succeeded in cloud adoption. The enterprises that moved fastest weren’t the ones with the best AI models; they were the ones that solved the governance and compliance problem first. For automation architects writing business cases, leading with governance and compliance is more likely to get executive sign-off than leading with productivity gains.

Customer Evidence: Rolls-Royce and Beyond

Rolls-Royce deployed Now Assist (internally branded “Merlin”) in August 2025 across 12,000 employees, processing over 10,000 conversations per month. The reported results: 5,000 hours of efficiency savings, a 54% deflection rate, and a two-day mean time to resolve. Plans call for rollout across all services by October 2026.

HDFC Bank’s group CIO described AI Control Tower as “the common governance layer” across the bank’s AI use cases — a significant validation from one of the world’s largest banks, where regulatory compliance is not optional but existential.

IDC projects active AI agents growing from roughly 28.6 million in 2025 to over 2.2 billion by 2030. ServiceNow’s security and risk business alone crossed $1 billion in annual contract value in 2025. These numbers frame the scale of the governance problem: billions of agents will need governance, and the vendors who solve it first will own the enterprise AI control plane.

Frequently Asked Questions

Is the ServiceNow MCP Server free for existing customers?

Yes. The MCP Server is included in every Now Assist and AI Native SKU at no additional cost. Headless actions consume the same Assist currency already used for Now Assist and AI Agents, so there’s no separate procurement process. Additional Action Fabric features are planned for the second half of 2026.

Can I use AI Control Tower to govern agents that don’t run on ServiceNow?

Yes, that’s the core design intent. AI Control Tower’s 30 integrations with AWS, Google Cloud, Azure, SAP, Oracle, and Workday allow it to discover, observe, and govern AI assets across platforms. The AI Gateway provides real-time governance for MCP transactions from third-party AI systems. However, the depth of governance depends on the integration — ServiceNow-native agents get the fullest governance stack, while external agents get discovery, monitoring, and identity governance.

How does ServiceNow’s approach differ from Salesforce Agentforce?

Salesforce’s Agentforce is optimized for CRM and customer-facing agent scenarios, built on top of Salesforce’s customer data graph. ServiceNow’s Action Fabric is designed for cross-functional enterprise operations — IT, HR, security, CRM, and more — built on the CMDB and workflow transaction history. Salesforce launched Headless 360 at TDX 2026 for a similar headless execution model, but ServiceNow’s governance layer (AI Control Tower) and security acquisitions (Armis, Veza) give it a broader cross-platform governance reach. In practice, many enterprises will run both.

What does the Armis acquisition mean for existing ServiceNow security customers?

Armis extends ServiceNow’s security scope from IT assets to OT, IoT, and connected devices — nearly 7 billion assets tracked globally. Its data feeds directly into the CMDB, creating a continuously updated view of the attack surface. Combined with Veza’s identity governance, the Autonomous Security and Risk specialists can handle vulnerability resolution and security operations end-to-end, with ServiceNow’s internal team reporting a 7x improvement in incident handling speed.

Is ServiceNow replacing RPA with AI agents?

Not directly. ServiceNow positions itself as the governance and execution layer, not as an RPA replacement. AI Control Tower’s cross-platform discovery includes RPA bots alongside AI agents — treating them as part of the same automation estate. The NVIDIA Project Arc partnership (desktop agents) does overlap with traditional attended RPA, but as an evolution toward autonomous desktop work rather than a direct replacement. Existing RPA investments can interoperate with ServiceNow through the MCP Server and the 30 new platform integrations.

Key Takeaways

• AI Control Tower is now a five-function governance platform (Discover, Observe, Govern, Secure, Measure) covering AI assets across ServiceNow and external platforms including AWS, Azure, Google Cloud, SAP, Oracle, and Workday.
• Action Fabric with its generally available MCP Server lets any AI agent — Claude, Copilot, custom-built — execute governed enterprise workflows headlessly, with full audit trails and identity verification.
• Autonomous Workforce AI specialists go beyond copilot-style assistance to complete end-to-end business processes in IT, CRM, and (soon) security, HR, and finance.
• $11+ billion in acquisitions (Armis, Veza, Traceloop, Moveworks) provide the technical foundation for security, identity governance, observability, and conversational AI.
• The core platform is GA now — Action Fabric, AI Control Tower, IT and CRM specialists, and Otto are all generally available, not roadmap promises.
• MCP as the enterprise agent standard gains further momentum with ServiceNow’s adoption, reinforcing the protocol’s position alongside Google’s A2A as the integration layer for multi-agent enterprise systems.
• NVIDIA Project Arc signals the convergence of autonomous desktop agents and traditional attended RPA, with ServiceNow governance and NVIDIA compute providing the production foundation.
• For RPA teams, the strategic question is not whether to adopt agentic AI governance but which platform’s governance layer to standardize on — and ServiceNow’s CMDB-rooted approach is now the most architecturally complete option available.

References

1. ServiceNow Newsroom, “ServiceNow expands AI Control Tower to discover, observe, govern, secure, and measure AI deployed across any system in the enterprise,” May 5, 2026. newsroom.servicenow.com
2. ServiceNow Newsroom, “ServiceNow opens its full system of action to every AI Agent in the enterprise,” May 5, 2026. newsroom.servicenow.com
3. Reworked, “ServiceNow Wants to Be the Control Layer for Every AI Agent in the Enterprise,” 2026. reworked.co
4. Diginomica, “ServiceNow Knowledge 2026 — AI Control Tower expands, Autonomous Workforce reaches every function,” 2026. diginomica.com
5. CX Foundation, “ServiceNow Knowledge 2026: The Top 10 Announcements,” 2026. cxfoundation.com
6. No Jitter, “ServiceNow AI Platform: Any agent, any system,” 2026. nojitter.com
7. ERP Today, “ServiceNow Repositions Around AI Security and Governance at Knowledge 2026,” 2026. erp.today
8. Kellton, “How Agentic AI Reshapes Workflows: ServiceNow Knowledge 2026,” 2026. kellton.com
9. NAND Research, “ServiceNow Knowledge 2026: The Agentic Enterprise Takes Shape,” 2026. nand-research.com
10. Constellation Research, “ServiceNow Knowledge 2026: AI Control Tower, Action Fabric, Autonomous Workforce and more,” 2026. constellationr.com