No Kill Switch? No Problem.

Tim Boswell
ServiceNow Employee

2 hours ago

At the Knowledge Conference in May, the AI Kill Switch was demonstrated. Several of our Security Operations customers and partners have inquired about this feature because they are concerned about Agentic Workflows or AI Agents that are misconfigured, or they are provided wrong privileges, or they malfunction for any number of reasons.

 

Unfortunately, the AI Kill Switch feature will not be available for another month or two, but those same customers and partners have told me that they cannot wait and need a solution now. So, in the meantime, I have provided some options for stopping or shutting down Agentic Workflows, AI Agents and AI Specialists:

 

Immediate / Reactive Controls:

 

  1. Cancel via the In-Product UI (Fastest, No Admin Required)

When an agentic workflow is running, a contextual side panel shows all workflows associated with the record. To cancel an active workflow, select the three-dots (more options) menu on the agentic workflow card and choose Cancel.

This is the first thing a fulfiller or agent should try — no admin role needed, works in both Workspaces and UI16.

 

  1. Kill the Active Transaction (Admin — Platform Level)

If the UI cancel doesn't work or the agent is stuck "In Progress," the underlying platform transaction can be terminated directly.

To kill a running AI Agent job in ServiceNow, use the official documentation for viewing and terminating active transactions. Specifically, look for and terminate jobs named "Update AIA agent messages" or "Update AIA tool messages" if they are currently running.

How to do it:

  • Navigate to: All > System Diagnostics > Active Transactions (or type sys_activity.list in the filter navigator)
  • Or go directly to [your-instance].service-now.com/cancel_my_transaction.do to cancel your own session's transaction

Reference: View and kill active transaction — ServiceNow Product Docs (Zurich)

 

  1. Cancel Stuck Executions via the sn_aia_execution_plan Table (Admin)

For executions that remain stuck in "In Progress" state:

Navigate to the sn_aia_execution_plan table, sort by the Created field (newest to oldest), filter by State = In Progress, and cancel the stuck executions.

This is the table-level kill switch for agentic execution plans and is recommended by ServiceNow's AI Center of Excellence team.

 

Stop Recurrence / Prevent Re-Triggering

 

  1. Deactivate the Trigger in AI Agent Studio (Admin)

To prevent the agent from firing again while you investigate:

Navigate to AI Agent Studio > Create and Manage, find your AI Agent or Agentic Workflow, go to the Trigger tab, and set the trigger to Inactive or delete it.

 

  1. Disable the Agent's Status Toggle (Admin)

In AI Agent Studio > Create and Manage, open the agent, go to Define Availability, and switch the Status toggle to Off. This prevents any new executions from starting.

Governance-Layer Controls (Proactive / Reactive)

 

  1. AI Control Tower — Pause or Disable an Agent Centrally

The AI Control Tower lets admins set boundaries on what AI agents can and cannot do, including data access limits and approval requirements for sensitive actions. Admins can pause or disable agents that exceed their defined scope — keeping autonomous AI from overstepping its bounds.

ServiceNow positions the AI Control Tower as a governance layer where agents are traceable every time, and stoppable every time — a direct response to the risk that agentic systems could take irreversible actions in financial or operational systems.

 

References:

 

  1. Now Assist Guardian — Block Malicious or Anomalous Agent Behavior

If the malfunction is driven by a prompt injection attack or unexpected LLM behavior:

Now Assist Guardian is a built-in guardrail layer covering Prompt Injection Protection and Offensive Content Detection. Admins can configure it in Block + Log mode to prevent injected content from executing, or enable Log only mode to monitor before enforcing blocks.

ServiceNow provides prompt injection protection via Now Assist Guardian, and grounded prompt templates tie prompts to platform data to increase response accuracy.

 

References:

0 Helpfuls
  • 45 Views
Version history
Last update:
2 hours ago
Updated by:
ServiceNow Employee
Contributors