The Security Question
Controllers worry about sending sensitive ERP data to AI platforms. Understanding what data is transmitted, how it's protected, platform security certifications, and comparison to current practices addresses legitimate security concerns.
Reality: AI agent security is comparable to or better than existing ERP integrations and email communications.
What Data Gets Transmitted
Data Sent to AI Platforms
Customer information:
- Name and contact details (phone, email)
- Company name
- Account status (current balance, overdue days)
Transaction details:
- Invoice numbers
- Amounts and dates
- Payment terms
- Due dates
Interaction history:
- Previous communications
- Payment commitments
- Notes from staff
What Does NOT Get Transmitted
Never sent to AI platforms:
- Credit card numbers or payment method details
- Bank account information
- Social security numbers or tax IDs
- Passwords or authentication credentials
- Full financial statements
- Confidential contract terms
Why: Not needed for exception handling. AI only receives minimum data required for conversation.
Data Transmission Security
Encryption in Transit
All data encrypted:
- TLS 1.2 or higher
- Industry-standard encryption (256-bit)
- Same security as banking websites
Path:
- ERP → Workflow platform (encrypted)
- Workflow platform → AI platform (encrypted)
- AI platform → Voice platform (encrypted)
Comparison: Standard email is often unencrypted. AI integration is more secure than typical email communication.
Encryption at Rest
Data storage:
- All platforms encrypt stored data
- Encryption at rest (AES-256 typically)
- Access controls limit who can view
- Audit logs track all access
Retention:
- Call recordings: 30-90 days typically
- Transcripts: 90 days to 1 year
- Outcome data: Permanently in ERP (not AI platforms)
Platform Security Certifications
Major AI Platforms
OpenAI:
- SOC 2 Type II certified
- Does not train on customer data
- Enterprise data retention policies
- GDPR and CCPA compliant
Anthropic (Claude):
- SOC 2 Type II certified
- Does not train on customer data
- Enterprise privacy commitments
- GDPR compliant
Google (Gemini):
- ISO 27001 certified
- SOC 2/SOC 3 compliant
- Does not train on customer data without permission
- GDPR compliant
Key point: Major platforms do not use customer conversation data for training AI models.
Voice Platforms
Twilio:
- SOC 2 Type II certified
- HIPAA compliant (BAA available)
- PCI DSS Level 1 certified
- ISO 27001 certified
- GDPR compliant
Vonage:
- SOC 2 certified
- ISO 27001 certified
- HIPAA compliant capabilities
- PCI DSS certified
Bandwidth:
- SOC 2 Type II certified
- HIPAA compliant
- Emergency services certified
Access Controls
Who Can Access Data
AI platform access:
- Your implementation partner (setup only)
- Platform provider staff (if support ticket opened)
- Automated systems only (no human review unless explicitly requested)
ERP data access:
- Service account with limited permissions
- Read access to customer and transaction data
- Write access to notes/tasks only (cannot modify transactions)
Call recordings:
- Your staff (full access)
- Implementation partner (if support needed, with permission)
- Platform provider (only if support ticket opened)
Role-Based Access Control
Implementation:
- Service account has minimum required permissions
- No access to payroll, HR, or financial reporting data
- Cannot delete or modify historical data
- Cannot access unrelated ERP modules
Comparison to staff: Staff often has broader ERP access than AI service account. AI access is more restricted.
Audit Trails
Complete Logging
Every AI interaction logged:
- Timestamp
- Customer contacted
- Data accessed
- Actions taken
- Outcome documented
Platform access logged:
- Who accessed system
- When access occurred
- What data was viewed
- Changes made (if any)
Compliance benefit: Better audit trail than manual processes. Complete documentation for compliance reviews.
Data Processing Agreements
Required Contracts
Data Processing Agreement (DPA):
- Defines how platforms handle your data
- Specifies security requirements
- Commits to compliance standards
- Required for GDPR compliance
Business Associate Agreement (BAA):
- Required for HIPAA compliance
- Available from major platforms
- Defines protected health information handling
- Necessary for healthcare implementations
Terms to require:
- No training on customer data
- Data deletion upon termination
- Encryption standards
- Breach notification procedures
- Audit rights
Comparison to Current Security
Email Communication Security
Current state:
- Most business email unencrypted in transit
- Email stored on multiple servers
- Forwarding creates copies outside control
- Difficult to audit who viewed
AI agent approach:
- All communications encrypted
- Centralized secure storage
- Complete audit trail
- Access controls enforced
Assessment: AI is more secure than standard email
Manual Phone Calls
Current state:
- Calls rarely recorded
- No encryption (standard phone network)
- No audit trail of what was said
- He-said-she-said disputes common
AI agent approach:
- All calls recorded
- Encrypted transmission (VoIP)
- Complete transcripts available
- Indisputable record of conversation
Assessment: AI provides better documentation and security
Spreadsheet Sharing
Current state:
- Customer lists in Excel
- Shared via email attachments
- Stored on local computers
- No encryption typically
- No access control once downloaded
AI agent approach:
- No spreadsheets shared externally
- Data accessed via secure API only
- Centralized access control
- Complete audit trail
Assessment: AI eliminates insecure spreadsheet sharing
Industry-Specific Considerations
Healthcare (HIPAA)
Requirements:
- Business Associate Agreement with all platforms
- Encryption in transit and at rest
- Access controls and audit logs
- Breach notification procedures
AI platform compliance:
- Major platforms offer HIPAA-compliant configurations
- BAAs available
- Additional cost may apply ($0-$500 monthly)
Suitability: AI agents can be HIPAA-compliant with proper platform selection and configuration
Financial Services (PCI DSS, FINRA)
Requirements:
- Never transmit card numbers or banking details
- Communication compliance (FINRA if applicable)
- Recordkeeping requirements
AI platform compliance:
- Voice platforms are PCI DSS certified
- Call recordings meet FINRA requirements
- No PCI data transmitted (not needed for collections)
Suitability: AI agents appropriate with proper data filtering
Government Contractors (DFARS, FedRAMP)
Requirements:
- May require FedRAMP authorized platforms
- Controlled Unclassified Information (CUI) handling
- Additional security controls
AI platform compliance:
- Few AI platforms currently FedRAMP authorized
- Azure OpenAI has government cloud options
- May require on-premise or government cloud deployment
Suitability: Limited options currently. Evaluate carefully for CUI/classified data.
Risk Mitigation Strategies
Data Minimization
Approach: Only transmit data absolutely necessary
Implementation:
- Filter out sensitive fields before transmission
- Use customer ID instead of full details where possible
- Summarize rather than transmit complete history
- Exclude data unrelated to exception handling
Geographic Data Residency
Requirement: Data must stay in specific country/region
Platform options:
- OpenAI: US, EU regions available
- Anthropic: US, EU regions
- Google: Extensive regional options
- Voice platforms: Regional deployment available
Cost impact: Often no additional cost, sometimes slight premium
Private/Dedicated Infrastructure
When needed: Highest security requirements
Options:
- Azure OpenAI (dedicated deployment)
- AWS Bedrock (Claude via AWS)
- Private workflow servers
Cost impact: Significantly higher ($2,000-$10,000+ monthly)
Suitable for: Enterprise scale, extreme security requirements
Breach Response
Platform Breach Scenarios
What happens if platform breached:
- Platform notifies customers per DPA (typically 72 hours)
- Investigation of exposure
- Remediation steps
- Customer notification if data exposed
Your response:
- Assess what data was potentially exposed
- Notify affected customers if required
- Consider switching platforms if confidence lost
- Review and strengthen security measures
Historical context: Major platforms (OpenAI, Google, Twilio) have strong security track records. Breaches rare.
The Reality
AI agents are safe for sensitive ERP data with proper controls. Major platforms (OpenAI, Anthropic, Google, Twilio) are SOC 2 certified, encrypt data in transit and at rest, do not train on customer data, provide complete audit trails.
Data transmitted: Customer contact info, account status, transaction details. Not transmitted: Payment methods, banking info, SSN/tax IDs.
Security often better than current practices (unencrypted email, unrecorded calls, spreadsheet sharing).
Industry-specific compliance available: HIPAA (BAA available), PCI DSS (platforms certified), FINRA (recordkeeping supported).
Risk mitigation through data minimization, geographic residency options, encryption standards, and contractual protections (DPA/BAA).
Comparison: AI security comparable to other ERP integrations, better than typical email and phone communications.