Zscaler — AI/ML application category monitoring
Zscaler Internet Access (ZIA) URL filtering policy to monitor and optionally block access to the AI/ML application URL category, providing visibility into Shadow AI usage across the organization.
Rule
# Zscaler Internet Access (ZIA) — URL Filtering Policy for AI/ML Category
# Navigation: ZIA Admin Portal > Policy > URL & Cloud App Control
#
# Policy rule: Monitor AI/ML Application Category
#
# Rule order: Place BEFORE general allow rules, AFTER sanctioned AI allow rules
#
# Criteria:
# - URL Category: 'AI/ML Applications' (Zscaler-managed category)
# - Additional custom category (if needed): 'Shadow AI — Custom'
# populated from /watchlists/ConsumerLLMDomains.csv
# - Users: All users (exclude service accounts and break-glass)
# - Departments: All (or phased by department for rollout)
# - Locations: All
# - Time: Always
#
# Action: Caution (displays warning page, user can proceed — logged)
# Alternative actions:
# - Allow (monitor only — log all access silently)
# - Block (full enforcement — use after comms rollout)
#
# Logging:
# - Enable full logging for this rule
# - Log fields: user, department, URL, category, action, bytes uploaded/downloaded
# - Forward logs to SIEM via Nanolog Streaming Service (NSS)
#
# Companion rule (higher priority):
# - Name: Allow Sanctioned AI Services
# - URL category: custom 'Sanctioned AI' (Azure OpenAI, enterprise Copilot endpoints)
# - Action: Allow
# - Ensures sanctioned tools are not blocked by the monitoring rule below# Zscaler Internet Access (ZIA) — URL Filtering Policy for AI/ML Category # Navigation: ZIA Admin Portal > Policy > URL & Cloud App Control # # Policy rule: Monitor AI/ML Application Category # # Rule order: Place BEFORE general allow rules, AFTER sanctioned AI allow rules # # Criteria: # - URL Category: 'AI/ML Applications' (Zscaler-managed category) # - Additional custom category (if needed): 'Shadow AI — Custom' # populated from /watchlists/ConsumerLLMDomains.csv # - Users: All users (exclude service accounts and break-glass) # - Departments: All (or phased by department for rollout) # - Locations: All # - Time: Always # # Action: Caution (displays warning page, user can proceed — logged) # Alternative actions: # - Allow (monitor only — log all access silently) # - Block (full enforcement — use after comms rollout) # # Logging: # - Enable full logging for this rule # - Log fields: user, department, URL, category, action, bytes uploaded/downloaded # - Forward logs to SIEM via Nanolog Streaming Service (NSS) # # Companion rule (higher priority): # - Name: Allow Sanctioned AI Services # - URL category: custom 'Sanctioned AI' (Azure OpenAI, enterprise Copilot endpoints) # - Action: Allow # - Ensures sanctioned tools are not blocked by the monitoring rule below
How it works
Zscaler maintains a curated 'AI/ML Applications' URL category that is continuously updated as new AI services emerge. This policy leverages that category to log, caution, or block access to AI/ML applications. The 'Caution' action is recommended for initial deployment — it presents a warning page to the user (with AUP reference) while still logging the access, providing both visibility and gentle deterrence without hard-blocking productivity. Pair with NSS log streaming to a SIEM for correlation with other Shadow AI detections.
Required data sources
- Zscaler Internet Access (ZIA) web logs
- Zscaler Client Connector deployed on endpoints (or GRE/IPsec tunnel from network)
Prerequisites
- Zscaler Internet Access (ZIA) licensed — Business or Transformation edition
- Zscaler Client Connector deployed to managed endpoints
- SSL/TLS inspection enabled for AI/ML category URLs
- Nanolog Streaming Service (NSS) configured for SIEM integration (optional but recommended)
Expected volume
Caution mode: 100-300 caution-page impressions per day per 10k users initially. Drops 60-70% within 2 weeks as users shift to sanctioned tools. Block mode: 30-80 block events per day at steady state.
False-positive guidance
Zscaler's AI/ML category may include enterprise-legitimate AI tools. Create an explicit allow rule for sanctioned AI services with higher priority. Occasionally, non-AI SaaS tools with AI features may be categorized under AI/ML — review the Zscaler URL lookup tool for miscategorizations and submit recategorization requests.
Tuning steps
- Start with Allow (log-only) for 1 week to baseline usage volume.
- Move to Caution for 2 weeks — monitor user response and override rates.
- Move to Block after communications rollout and sanctioned-tool availability confirmed.
- Create custom category 'Sanctioned AI' for enterprise-approved AI services and place allow rule above this policy.
- Review Zscaler category updates monthly — new AI services are added continuously.
Framework mappings
- NIST AI RMF: GOVERN-4.1
- ISO/IEC 42001: 8.2
- NIST CSF: DE.CM-1, DE.CM-7