Netskope — Real-time DLP for AI service uploads
Data Loss Prevention policy in Netskope that inspects content uploaded to AI services in real time, detecting and blocking sensitive data exfiltration through generative AI platforms.
Rule
# Netskope — Real-time DLP policy for AI service uploads
# Navigation: Policies > Real-time Protection > New Policy
#
# Policy: AI Service DLP — Block Sensitive Uploads
#
# Source:
# - Users: All users (exclude break-glass / service accounts)
# - User groups: All (or phased rollout by department)
#
# Destination:
# - Application tag: 'Generative AI', 'AI/ML', 'AI Code Assistant'
# - App instance: 'Unsanctioned' OR 'Monitored'
# - Specific apps (if not using tags): ChatGPT, Claude, Gemini, Perplexity, DeepSeek
#
# Activity:
# - Upload, Post, Send, Share, Create
#
# DLP Profile (Settings > DLP > Profiles):
# - Profile name: AI-Upload-Sensitive-Content
# - Rules (match any):
# - PII: SSN, credit card, passport, national ID, email + phone combo
# - PHI: medical record numbers, ICD/CPT codes, HIPAA identifiers
# - Financial: IBAN, SWIFT, bank account patterns
# - Source code: code file extensions (.py, .js, .java, .cs, .go, .rs, .sql)
# - Proprietary: custom regex for internal project codenames, customer IDs
# - Classification labels: files tagged Confidential, Highly Confidential, Restricted
# - Severity: Critical for PII/PHI, High for source code, Medium for proprietary
#
# Actions:
# - Block with user notification (justification required for override)
# - Alert: High severity
# - Incident: create in Netskope Incident Management
# - Notification: email to AI Governance team
# - Log: full activity detail in SkopeIT
#
# Exceptions:
# - Sanctioned enterprise AI instances (e.g., Azure OpenAI with corporate tenant)
# - Users in approved AI pilot program (sync group from IdP)# Netskope — Real-time DLP policy for AI service uploads # Navigation: Policies > Real-time Protection > New Policy # # Policy: AI Service DLP — Block Sensitive Uploads # # Source: # - Users: All users (exclude break-glass / service accounts) # - User groups: All (or phased rollout by department) # # Destination: # - Application tag: 'Generative AI', 'AI/ML', 'AI Code Assistant' # - App instance: 'Unsanctioned' OR 'Monitored' # - Specific apps (if not using tags): ChatGPT, Claude, Gemini, Perplexity, DeepSeek # # Activity: # - Upload, Post, Send, Share, Create # # DLP Profile (Settings > DLP > Profiles): # - Profile name: AI-Upload-Sensitive-Content # - Rules (match any): # - PII: SSN, credit card, passport, national ID, email + phone combo # - PHI: medical record numbers, ICD/CPT codes, HIPAA identifiers # - Financial: IBAN, SWIFT, bank account patterns # - Source code: code file extensions (.py, .js, .java, .cs, .go, .rs, .sql) # - Proprietary: custom regex for internal project codenames, customer IDs # - Classification labels: files tagged Confidential, Highly Confidential, Restricted # - Severity: Critical for PII/PHI, High for source code, Medium for proprietary # # Actions: # - Block with user notification (justification required for override) # - Alert: High severity # - Incident: create in Netskope Incident Management # - Notification: email to AI Governance team # - Log: full activity detail in SkopeIT # # Exceptions: # - Sanctioned enterprise AI instances (e.g., Azure OpenAI with corporate tenant) # - Users in approved AI pilot program (sync group from IdP)
How it works
Netskope's inline DLP engine inspects content in real time as it passes through the Netskope proxy. This policy targets uploads, posts, and shares to AI-tagged applications, applying a comprehensive DLP profile that checks for PII, PHI, financial data, source code, and proprietary content. Unlike endpoint-only DLP, Netskope inspects traffic regardless of the browser used, covering scenarios that endpoint DLP may miss (e.g., API-based uploads, non-Edge browsers without extensions). The block-with-justification model balances security with productivity.
Required data sources
- Netskope Intelligent SSE with DLP add-on
- Netskope Client deployed for inline inspection
- SSL/TLS inspection enabled for AI service domains
Prerequisites
- Netskope Intelligent SSE with Advanced DLP licensed
- SSL/TLS decryption enabled and CA certificate deployed to endpoints
- Netskope Client deployed and steering configured for web traffic
- DLP profiles configured with organization-specific sensitive info patterns
- AI application tags populated in the CCI catalog (see DET-NS-001)
Expected volume
After initial tuning: 10-50 DLP incidents per day per 10k users. Expect higher initial volumes during the first 2 weeks before user awareness improves.
False-positive guidance
Common FP sources: (1) sample data in training documents containing realistic-looking PII patterns, (2) code snippets in documentation that match source-code rules, (3) public datasets containing synthetic PII. Tune DLP fingerprinting thresholds and add exact-match exceptions for known benign patterns.
Tuning steps
- Deploy in alert-only mode for 2 weeks to baseline DLP match volume and FP rate.
- Refine DLP profile rules based on FP analysis — adjust confidence thresholds and minimum match counts.
- Enable block-with-override after tuning; review override justifications weekly.
- Escalate to block-without-override for highest-severity SITs (PII, PHI) after 30 days of stable operation.
- Exclude sanctioned enterprise AI instances from the block policy.
Framework mappings
- NIST AI RMF: GOVERN-4.1, MEASURE-2.7, MANAGE-2.2
- ISO/IEC 42001: 8.2, 8.3, 8.4
- MITRE ATT&CK: T1567.002
- NIST CSF: DE.CM-3, PR.DS-5