Kill Switch in Now Assist AI Agents

  • Release version: Zurich
  • Updated June 4, 2026
  • 3 minutes to read
    • Summarize
    Summarized using AI
    This content was generated using new OpenAI-powered functionality. Results are provided on an as is basis and are not guaranteed to be accurate or complete.

    Summary of Kill Switch in Now Assist AI Agents

    The Kill Switch feature in Now Assist AI Agents is designed to detect and stop runaway AI agent triggers that repeatedly activate on the same records. This prevents unnecessary consumption of Now Assist interactions, which can lead to increased costs without customer value. The feature monitors trigger activity and automatically disables an agent if a predefined threshold of repeated activations is exceeded.

    Show full answer Show less

    Key Features

    • Detection of Repeated Triggering: Identifies when the same record triggers the same AI agent objective more than a set number of times within a 24-hour period.
    • Configurable Thresholds: Default thresholds include 5 fires per record in 24 hours, 25 distinct records breaching the threshold, and 3 consecutive days of breach before action. These can be customized via system properties.
    • Two Execution Paths:
      • Trigger Activation Path: Records each trigger fire without interrupting conversations.
      • Evaluator Path: A daily scheduled job analyzes audit data, sends tiered warning emails, and disables triggers if necessary.
    • Operating Modes Controlled by System Property:
      • off – Auditing only, no warnings or disabling.
      • warnonly – Sends warnings without disabling (default mode).
      • enforce – Sends warnings and disables triggers after sustained breaches.
    • Alert Notifications: Warning emails are sent to trigger owners and associated agent or workflow creators with details and direct links to the trigger configuration.

    Practical Impact for ServiceNow Customers

    This feature helps customers avoid unexpected costs and inefficiencies caused by misconfigured AI agent triggers firing excessively. By monitoring and controlling trigger behavior, the Kill Switch ensures AI agents operate effectively without consuming unnecessary assists. Customers can tune thresholds and operating modes to fit their environments and receive proactive alerts to address potential issues promptly.

    Re-enabling a disabled trigger resets the monitoring cycle, allowing customers to manage and recover from false positives or resolved issues smoothly.

    The kill switch feature detects and stops runaway AI agent triggers that execute repeatedly against the same records, preventing unnecessary consumption of Now Assist interactions.

    Kill Switch Overview

    An AI agent trigger can match the same record multiple times within a short period during executions. When left unchecked, a misconfiguration trigger condition can process the same incident five or more times per day, across dozens of records, for several consecutive days. Each trigger activation spawns a conversation and consumes assists, resulting in charges for activity that provides no customer value.

    The kill switch feature detects when the same record is triggering the same agent objective beyond a threshold in a single day and automatically disables the agent. It monitors trigger activity, issues tiered alert notifications to trigger and agent owners, and optionally disables the trigger automatically after a configurable breach threshold is reached.

    Default detection thresholds

    The kill switch evaluates trigger activity against the following default thresholds. All thresholds are configurable by users through system properties:

    • Fires per record per 24-hour window: 5
    • Distinct records breaching the threshold: 25
    • Consecutive days of breach before action: 3
    Five tunable system properties control these thresholds and the feature's operating mode:
    • kill_switch.mode: Default value: warn_only. For the different operating modes the property contains, see Operating modes.
    • kill_switch.max_fires_per_window: Fires per record that mark it as breaching. Default value: 5.
    • kill_switch.min_distinct_records: Breaching records needed for the window to count as runaway. Default value: 25.
    • kill_switch.window_size: Length of one observation window. Default value: 1440 min / 24h.
    • kill_switch.consecutive_windows_duration: The total look back span. Default value: 4320 min / 3 days.

    Architecture

    The kill switch uses two fully decoupled execution paths.

    Trigger activation path
    Runs on every conversation start. When a trigger matches a record and fires, an audit row is written to the audit table before the conversation begins. Audit writes never interrupt or fail the user's conversation.
    Evaluator path
    A scheduled job runs once daily. It reads the audit table, computes a cycleStart date, and assigns each active trigger a stage value (K) from 0 to 3 based on the number of consecutive days the trigger has breached the threshold. If K is 1 or higher, the evaluator sends a warning email. If K reaches 3 and enforce mode is active, the evaluator also disables the trigger.
    Note:
    • Warning emails are sent to the trigger creator and to the creators of any agents or workflows associated with the trigger. Alerts follow the same notification mechanism used for all existing Now Assist alert purposes.
    • Re-enabling an inactive trigger resets the cycle, so the next evaluator run treats the trigger as a fresh Day 1.

    Operating modes

    The kill_switch.mode system property controls how the feature responds to a detected breach.
    off
    Audit rows are still written, but the evaluator performs no detection, sends no warnings, and never disables triggers.
    warn_only
    The evaluator sends a daily warning email for each day the breach pattern is present. Triggers are never inactive. This is the shipped default.
    enforce
    The evaluator sends the same tiered warnings on Days 1 and 2. On Day 3, it sends a final warning and deactivates the trigger configuration, including associated many-to-many and override rows.

    Escalation sequence

    In warn_only or enforce mode, the evaluator sends progressively stronger email notifications as the breach continues.

    • Day 1 — Early warning: [1 of 3] Action Recommended — Trigger Firing at Unusual Rate
    • Day 2 — Stronger warning: [2 of 3] Action Required — Trigger Has Been Firing at High Volume for 2 Consecutive Days
    • Day 3 — Final warning (enforce mode also disables): [3 of 3] Trigger inactive — High Volume Detected for 3 Consecutive Days

    Each email includes the agent name, trigger name, affected record count, and a direct link to the trigger configuration.