AI Control Tower release notes

topic brazil Updated Sep 10, 2026 16 min read ai-control-tower-rn.md
Raw file servicenow.com

The ServiceNow® AI Control Tower application provides a centralized workspace to track and act on AI governance work across the enterprise. See the following sections for release notes by version.

About AI Control Tower

  • Discover and maintain a complete, current inventory of AI systems, models, datasets, and prompts across ServiceNow and external platforms.
  • Track regulatory compliance posture against frameworks like NIST AI RMF and EU AI Act across your entire AI portfolio.
  • Secure AI systems across platforms with automated scoring and access hygiene, model vulnerability and validation findings, real-time guardrails against prompt injection and data leakage, and graduated containment for AI agents.
  • Evaluate the quality and safety of AI interactions across your portfolio using automated scoring, configurable metrics, and trend analysis for both ServiceNow and external AI systems.
  • Quantify the business value of AI investments with productivity, cost savings, and adoption metrics tied to individual AI systems.
  • Coordinate governance work across cross-functional teams with lifecycle playbooks, approval routing, and AI-generated recommendations.

See AI Control Tower for more information.

Activation and other requirements

Important: AI Risk and Compliance workstream activities, including impact assessments, risk assessments, control attestations, issues, policy exceptions, AI cases, and AI inquiries, aren't supported in the AI-native experience. This also applies to business configuration and administration activities, such as setting up risk assessment methodologies and assessment templates. Access and view the system of record through the workspace-based experience.

Accessibility and localization

  • Accessibility information

    • AI asset security score details, AI agent containment, and configuration pages on the Security tab are fully navigable by keyboard and compatible with screen readers.

  • Localization information

    Strings in AI asset security score details, AI agent containment, and configuration pages on the Security tab are available in your organization's configured language, including strings that were previously untranslated.

Parent Topic:AI Experiences release notes

Brazil Patch 2

The Brazil Patch 2 release adds improvements for policy management, security, discovery, and client-based access controls for AI Gateway.

What's new

  • ServiceNow Arc policy management

    Create policies and rules that control ServiceNow Arc usage.

  • AI Gateway client-based access policies

    Create client access policies that grant each MCP client access to specific tools on an MCP server. When you turn on access management for the server, clients can use only the tools that their policies grant.

  • Security settings usability improvements and AI usage consent

    Configure security settings using new sub-tabs in security settings: Overview, Runtime, Post-runtime, and Security insights. These sub-tabs correspond to metrics they configure, except Security insights. Security insights shows the status of the AICT Security Analyzer agent. In the Runtime sub-tab, the link to configure data patterns changed to go to Anonymization policies in Data Privacy. In the Post-runtime and Security insights sub-tabs, you're prompted to consent to consumption of assists when enabling settings.

  • Agent containment list filter and rename

    Filter the list of contained or reinstated agents by a particular agent. The Contained agents list is renamed to Agent containment list.

  • Post-runtime metrics configuration change and category rename

    When you select Configure, you're directed to Settings to update your configuration instead of to a side panel. AICT correctness detection is renamed to Correctness detection.

What's deprecated or removed

  • V1 AI asset onboarding playbook

    Starting with the Brazil Early Availability release, the V1 AI asset onboarding playbook no longer receives updates and isn't active for new customers. New customers use the V2 AI asset onboarding playbook. Existing customers who use or have customized the V1 playbook can continue to use it or migrate to the V2 playbook.

    For details, see the Deprecation Process [KB0867184] article in the Now Support Knowledge Base.

  • Prompt injection settings removed

    In Settings, under ServiceNow AI systems, the max skill calls and sample rate settings are removed from prompt injection. Instead, adjust these settings in AI Guardian analytics.

Brazil Patch 1

The Brazil Patch 1 release adds improvements for policy management, improved monitoring visualizations, and more.

What's new

  • Monitor agent activity chart improvements

    Monitor agentic AI performance over time using new filter options in the Monitor agent activity chart. View the top five lowest performing metrics, top five highest performing metrics, or view performance for a specific metric in the chart.

  • Session details improvements

    View a list of lowest scoring metrics and the LLM judge's reasoning in the Quality and Safety score cards on the Session details page.

  • Connectors

    Amazon Quick connector is a new connector which is part of discovering Systems, models, and prompts for creating AI connections.

  • Edit an Explicit Block policy

    Change who an Explicit Block policy blocks, what they're blocked from using, or its follow-up actions after the policy is published.

  • Block AI usage by department

    Scope an Explicit Block policy to a department so everyone in that department is blocked from a specific AI agent, model, or domain.

  • Use new role to contain AI agents

    Perform AI agent containment (kill switch) with the AI Security Operator role. You can deactivate an AI agent without requiring full AI Steward–level access.

  • Check whether the Security Analyzer agent is enabled

    Check the status of the Security Analyzer agent which determines security event severity and insights for your top recommendations, data integrity incident detection, system prompt leakage, correctness detection, prompt injection, and agent goal deviation metrics. Available on the Security Insights tab in Settings > Rules and templates > Security.

  • Enterprise Architecture for AICT plugin

    • Associate AI systems with business applications whether you use Enterprise Architecture Workspace. The new Enterprise Architecture for AICT plugin stores these associations in its own data model and installs automatically as a dependency of AI Control Tower Core (app-ai-governance) application.
    • Select business applications while creating an AI system in AI Control Tower. View AI adoption across your business application portfolio on the new Business portfolio tab in the AI Control Tower workspace homepage. For more information, see AI Control Tower integration with Enterprise Architecture>.

  • Cost types and sub-vendor rates

    Set a rate for a specific service from an integrated vendor, such as an agent platform. The sub-vendor rate takes precedence over the vendor rate. The vendor rate applies to any usage that doesn't have a sub-vendor rate.

  • Subscription-based costs for AI vendors

    Record per-seat subscription costs for integrated and non-integrated vendors. Specify the number of licensed seats, the cost per seat, and the contract dates. Optionally, add a one-time setup fee and distribute it over a set number of months so that it's included in the estimated monthly cost.

  • AI Connector for Amazon

    A new source system Amazon Quick has been added to the AI Connector for Amazon.

  • AI Connector for Anthropic

    Configure the Anthropic AI connection to opt in to discovery with beta APIs.

    The October version of the connector supports MID Server.

  • AI Connector for Salesforce

    Establish a sever-to-server integration without manual intervention with JWT connection.

  • AI Connector for OpenAI

    Configure mTLS certificate-based authentication besides the API Key for better security.

    The October version of the connector supports X.509 certificate creation.

  • AI Connector for Microsoft - Copilot Lite

    You can n w discover AI agents from Copilot Lite (Builder Agent) following a set of prerequisites on the Microsoft Entra admin center.

  • AI Connector for Google Enterprise - Agent Platform

    Configure the Anthropic AI connection to opt in to discovery with beta APIs.

    The October version of the connector supports MID Server.

What's changed

  • View input and output in trace details

    Toggle between input and output when viewing trace details.

  • Usability improvements to Security Overview metrics

    See only actionable items in your top recommendations on the Security Overview tab. In addition, the recommendations rank AI agent insights by number of critical security events and show the agent name, critical event count, and top threat categories. The security events list now matches the list on the Post-runtime tab. The Access issues detailed view includes a description that explains each access issue in plain language—identifying the agent, the operation, the resource, and the denial count.

  • Domain separation and AI Control Tower

    Review sensitive data metrics for ServiceNow AI systems in a domain-separated instance. Available on the Runtime tab in Security.

  • Configure Excessive Agency in Security

    Control whether Excessive Agency is enabled in Post-runtime configuration in Security. This setting controls data for Access issues and Privileged AI agents metrics, as well as post-runtime metrics. The setting is off by default.

  • Post-runtime Security probabilistic metrics and AI agent disabled by default

    To reduce token consumption, screening for data integrity incident detection, system prompt leakage, correctness detection, prompt injection, and agent goal deviation is disabled by default. In addition, the Security Analyzer agent that determines security event severity and insights is disabled by default. The default sampling rate for all metrics is 1%. If you're upgrading, your Detection enabled setting for each metric isn’t affected. After upgrading, check your settings in Settings > Rules and templates > Security and adjust if needed.

What's deprecated or removed

  • Now LLM Service deprecation notice

    Starting with the September 2026 release, Gemma 4 joins our growing portfolio of open-weight models available through Now LLM Service. The latest industry advancements are available alongside sovereignty-focused options. All models are hosted and governed by ServiceNow with the same infrastructure and data protections. Older models will remain available for existing published AI skills, agents, and agentic workflows, but will no longer be available for new development or configuration. For details, see the KB3066214: ServiceNow Otto Model Upgrades article in the Now Support Knowledge Base.

  • Traceloop Warehouse HTTPS connection removed

    The base system Traceloop Warehouse HTTPS connection (sn_ai_security.Traceloop_API alias) is removed. If you're upgrading and use Traceloop-based external AI metrics, create a new HTTPS connection under that alias and re-create your credential with connection URL https://api.traceloop.com.

  • Data integrity incident detection categories removed from external AI setting

    All categories are removed. For example, Guardrail Circumvention Attempt and Untrusted Links and Downloads are unavailable in the external AI setting.

Brazil Early Availability

The Brazil Early Availability release adds support for domain separation, policies, quality and safety metrics for specific assets, and more.

What's new

  • Control AI asset usage through policies

    Create AI asset usage policies that automatically respond to detected AI threats or block an AI agent, domain, or model outright.

  • Add an Azure AI Foundry connection

    Connect Azure AI Foundry to AI Control Tower so that policies and AI agent containment using kill switch protocol can reach and act on agents running on Azure AI Foundry.

  • Add metrics from the AI system record

    Add evaluation metrics for a specific AI system without changing your organization's global metric configuration.

  • Add specific metrics for one or more external AI systems

    Add, remove, or adjust the sample rate of metrics for one or more external AI systems, without changing your organization's global metric configuration.

  • Add specific metrics for one or more ServiceNow AI systems

    Add or remove metrics for one or more ServiceNow AI systems, without changing your organization's global metric configuration.

  • Exclude an AI system from a metric

    Exclude one or more AI systems from a specific metric, without changing that metric's configuration for every other system.

  • AI system performance metrics

    Monitor operational health at a glance, including average latency and token usage per session.

  • Latency and span counts for sessions

    See each session's response time and span count in the evaluated sessions lists and on individual session detail pages.

  • Scoring bias indicator

    Learn when a composite score might be skewed. See how many evaluations back each metric, and identify when uneven evaluation coverage is influencing the score more than the configured weight suggests.

  • Trace data retention controls

    Keep session, trace, and span data for up to 30 days for scoring, or discard it to reduce storage usage, with quality and safety scores staying available either way. Discarding also disables AI Skill Kit insights and hides the evaluated sessions views.

  • Custom date ranges for monitoring

    Analyze exactly the period you need by specifying an exact start and end date, up to 18 months apart.

  • Review AI security posture in design-time metrics

    Use AI security posture metrics to identify potential configuration issues in AI agents, tools, MCP servers, and system prompts that could lead to security risk.

  • AI Inventory Intelligence Agent

    AI Inventory Intelligence Agent automatically analyzes the AI asset inventory, identifies assets with incomplete metadata, and generates enrichment recommendations to improve data quality and governance readiness.

  • Configuring connectors

    • The Microsoft connector introduces the A365 agent platform to discover and import AI assets into ServiceNow AI Control Tower.

  • Client registration and AI Gateway setup

    • The MCP and CIMD registered clients can be edited to update their configuration from the AI Gateway Setup tab.

  • Updated playbooks for governing managed assets

    Manage AI assets through structured life cycle workflows using the updated playbooks. These playbooks automate governance tasks, route approvals, and track compliance requirements across onboarding, maintenance, and retirement phases. Use the Review and switch button to review the effect on your existing playbooks, flows, and subflows before switching to the new playbook.

  • Role based data visibility

    AI Control Tower now scopes value, engagement, and cost data based on the user’s role and assigned AI systems.

    AI stewards can view data for all AI systems in an instance. Product owners can view data only for the AI systems that they manage.

    Data scope is determined by the Managed by field on the AI system record.

    Product owners can review and adjust measurements for the AI systems that they manage. Cost configuration remains read-only because vendor-level changes can affect AI systems outside a product owner’s scope.

What's changed

  • Domain separation and AI Control Tower

    Use AI Control Tower on a domain-separated instance. In Security, detail pages for Privileged AI agents, Dormant AI agents, Access issues, Security events detected, Agent map, Post-runtime, and AI asset security score include a Domain column that shows the domain the AI asset belongs to, or shows global or is empty if domain separation isn't configured. Your top recommendations is replaced by a Security events detected section on the Security dashboard, and Sensitive data metrics aren't shown for ServiceNow AI systems.

  • Governing AI asset security

    Agent status is renamed to Access posture on the Security tab for an AI asset. Also, Detection time is now the first column in the list of events.

  • Security design-time metrics reorganized into subtabs

    Design-time metrics are organized into three subtabs: AI security posture, AI vulnerabilities, and AI validation. If the AI Security Exposure Management plugin isn't installed, use the source dropdown to filter the metrics by source (for example, Cisco or HiddenLayer).

  • Governing AI asset security

    On the Security tab for an AI asset, access issues and access posture metrics aren't shown for AI models because they're agent-level signals derived from agent identity and permissions. Therefore, they don't apply to AI models.

  • Contain AI agents manually using kill switch protocol

    AI agent containment (kill switch) supports retrying a failed or partial containment operation from the asset record or the agent containment list. You can also deactivate any managed AI agent directly from its asset record, whether or not it has an associated security event. AI agent containment (kill switch) is now available directly from an AI asset's Overview tab, not just its Security tab, for security events of any severity.

  • Configuring security connections

    The Security tab under Settings > Integrations is renamed to Control Enforcement Points.

  • Add a Gemini Enterprise Agent Platform connection

    The GCP Vertex AI security connector is renamed to Gemini Enterprise Agent Platform.

  • Agent containment list

    The Kill Switch Protocol Log is renamed the Agent containment list, and the View details option on the containment banner is renamed View containment options. The list now includes Domain and Actions columns. Containment details now show how the containment was initiated (Manual or Automated) and identity and enforcement details. The list can be filtered using the All, In progress, or Contained options, which replace the previous Show all link.

  • Security dates and times shown in UTC

    Date and time values in the Security tab — including the agent containment list, the Overview tab's Dormant AI agents and Privileged AI agents details, and the Post-runtime tab's security events — are now shown in Coordinated Universal Time (UTC) rather than your local timezone.

  • AI Gateway settings

    Starting in the September 2026 release, AI Gateway is available in AI Control Tower.

  • Configuring connectors

    • The GCP Vertex AI connector is renamed to Gemini Enterprise Agent Platform.
    • The application AI Service Graph Connector for GCP is renamed to AI Service Graph Connector for Google.
    • The Salesforce connector is renamed to AI Connector for Salesforce.

  • Veza connector authentication

    The Veza connector now authenticates using OAuth 2.0. Configure an OAuth profile and alias on the platform, then select it from Settings > Integrations > Connectors. Depending on which OAuth credential is active, tokens are either issued per user and tied to user identity, or shared at the system level, replacing API key authentication as the default method. The Veza tile appears in Settings > Integrations > Control Enforcement Points.

  • Create Asset Task action renamed

    The Create AI Task action, available on AI asset security events, privileged AI agents, and dormant AI agents, is renamed to Create Asset Task.

  • AI security posture metrics added and renamed

    On the Design-time tab in Security, AI security posture subtab, three metrics were added under AI assets with critical findings: MCP servers, Tools, and System prompts.

    Several metrics on the same subtab were renamed for clarity:

    • Critical findings approaching remediation target is renamed Findings approaching remediation target.
    • Critical findings with remediation overdue is renamed Findings with remediation overdue.
    • Number of findings deferred is renamed Findings deferred.
    • Findings by risk rating is renamed Posture findings by risk rating.
    • Findings by OWASP LLM Top 10 is renamed Critical findings by OWASP top ten.
    • Findings by MITRE ATLAS techniques is renamed Critical findings by MITRE ATLAS technique. The Critical findings by platform, Critical findings by OWASP top ten, and Critical findings by MITRE ATLAS technique metrics require the AI Security Exposure Management plugin, and are no longer available in the base system.

  • Updated default Sampling rate and Max skill calls

    In Settings > Rules and templates > Security, the default values for Sampling rate and Max skill calls have changed to help ensure efficient, predictable analysis. If you're upgrading, your Sampling rate and Max skill calls are updated to the new defaults. Your Detection enabled setting for each metric is not affected. After upgrading, check your Sampling rate and Max skill calls settings under Settings > Rules and templates > Security and adjust if needed.

What's deprecated or removed

  • Now LLM Service deprecation notice

    Starting with the September 2026 release, Gemma 4 joins our growing portfolio of open-weight models available through Now LLM Service. The latest industry advancements are available alongside sovereignty-focused options. All models are hosted and governed by ServiceNow with the same infrastructure and data protections. Older models will remain available for existing published AI skills, agents, and agentic workflows, but will no longer be available for new development or configuration. For details, see the KB3066214: ServiceNow Otto Model Upgrades article in the Now Support Knowledge Base.

  • Veza API key authentication deprecated

    The Veza connector now authenticates using OAuth 2.0 instead of an API key. The sn_ai_security.veza.api.key system property is deprecated.

  • Findings by severity metric removed from Security

    The Findings by severity metric is removed from the Security Design-time tab, AI security posture subtab.

  • V1 AI asset onboarding playbook

    The V1 AI asset onboarding playbook no longer receives updates and isn't active for new customers. New customers use the V2 AI asset onboarding playbook. Existing customers who use or customized the V1 playbook can continue to use it or migrate to the V2 playbook.

    For details, see the Deprecation Process [KB0867184] article in the Now Support Knowledge Base.

Plugin information

  • New plugins

    AI Policy Framework (sn_ai_policy_framework): Mitigate AI exposure through policies.

Search documentation
Search document titles and file paths on the brazil branch