The email from legal arrived at 2:47 PM on a Tuesday. An associate had uploaded a client's merger documents to ChatGPT to help draft a fairness opinion. The documents contained material non-public information, valuation data, and the names of all parties involved. The associate used the free version with a personal account.
This scenario, or variations of it, plays out across industries. Employees discover that AI tools dramatically improve their productivity. They start using whatever is available. By the time compliance discovers the exposure, months of sensitive data may have been transmitted to external systems.
The enterprise version of ChatGPT exists precisely to address this problem. But enterprise features only matter if compliance officers understand what those features actually provide, where gaps remain, and what additional controls are necessary.
The short version: If you need to redact sensitive documents before they reach AI systems, PaperVeil handles that layer. The rest of this article explains where it fits in the broader governance architecture.
The Compliance Officer's Perspective
Compliance officers evaluating ChatGPT Enterprise need answers to specific questions:
What regulatory frameworks apply? Most organizations operate under multiple overlapping requirements: GDPR for EU data, CCPA/CPRA for California residents, HIPAA for health information, GLBA for financial data, SOX for public companies, industry-specific regulations, and state-level data protection laws.
What certifications demonstrate control effectiveness? Third-party attestations provide evidence that security controls operate as designed.
What contractual protections exist? Data processing agreements, business associate agreements, and enterprise contracts allocate responsibility and establish obligations.
What data handling practices create or mitigate risk? Training on customer data, retention periods, access controls, and breach notification procedures all affect exposure.
What monitoring and audit capabilities exist? You can't govern what you can't see.
ChatGPT Enterprise addresses many of these concerns. Understanding exactly what it addresses, and what it doesn't, is the compliance officer's job.
ChatGPT Enterprise Security Model
OpenAI's most recent SOC 2 Type 2 report covers January 1, 2025 through June 30, 2025 and is available through their Trust Portal. The report covers controls relevant to Security, Availability, Confidentiality, and Privacy Trust Services Criteria for ChatGPT Enterprise, ChatGPT Edu, ChatGPT Team, and the API Platform.
The key certifications include:
SOC 2 Type 2: Independently audited controls for security, availability, confidentiality, and privacy. This is the standard enterprise software assurance framework.
ISO 27001: Information security management system certification demonstrating systematic approach to protecting information assets.
ISO 27017: Cloud-specific security controls extending ISO 27001 for cloud service providers.
ISO 27018: Protection of personally identifiable information in public clouds.
ISO 27701: Privacy information management system extending ISO 27001 for privacy management.
CSA STAR: Cloud Security Alliance certification addressing cloud-specific security concerns.
These certifications indicate that OpenAI has implemented and maintains security controls that an independent auditor has validated. They don't guarantee security, but they provide evidence of mature security practices.
Data Handling
The most important commitment for compliance purposes: by default, OpenAI does not use data from ChatGPT Enterprise, ChatGPT Business, ChatGPT Edu, ChatGPT for Healthcare, ChatGPT for Teachers, or the API platform for training or improving models. This applies to both inputs and outputs.
This is fundamentally different from consumer ChatGPT, where user data may be used for training unless explicitly opted out.
Encryption
ChatGPT Enterprise uses AES-256 encryption for data at rest and TLS 1.2 or higher for data in transit. With Enterprise Key Management (EKM), customers can control their own encryption keys, providing additional control and compliance evidence.
Data Residency
Enterprise customers can choose data residency in multiple regions: Europe, United Kingdom, United States, Canada, Japan, South Korea, Singapore, India, Australia, and the United Arab Emirates. Regional data residency supports GDPR and other jurisdiction-specific requirements.
Contractual Frameworks
Data Processing Addendum (DPA)
OpenAI offers a Data Processing Addendum for ChatGPT Business, ChatGPT Enterprise, and API customers. The DPA clarifies roles and responsibilities under GDPR and other privacy regulations. It establishes:
- OpenAI's role as data processor
- Customer's role as data controller
- Permitted processing purposes
- Security obligations
- Sub-processor disclosure
- Breach notification requirements
- Data subject rights support
The DPA is essential for GDPR compliance when processing EU personal data. Without it, the transfer may lack adequate legal basis.
Business Associate Agreement (BAA)
For healthcare customers, OpenAI offers a Business Associate Agreement to support HIPAA compliance. The BAA is available for ChatGPT for Healthcare and API healthcare customers. It establishes HIPAA-required protections for protected health information.
Note that the BAA doesn't make ChatGPT HIPAA compliant by itself. It creates the contractual framework for compliant use. The customer remains responsible for appropriate use, access controls, and ensuring PHI is handled appropriately.
Where Enterprise Falls Short
Enterprise features address many compliance concerns, but gaps remain.
Content Risks
Enterprise controls protect data handling. They don't address what employees submit or what responses they receive.
A compliance team member could ask ChatGPT about regulatory interpretations and receive incorrect guidance. A financial analyst could receive hallucinated market data. An HR professional could receive advice that creates employment law exposure.
The AI's outputs carry no certification or assurance. Enterprise features protect the data pipeline. They don't validate the content.
Third-Party Risk
Even with enterprise controls, you're still transmitting sensitive data to a third party. OpenAI's security practices are audited, but breaches happen to audited companies. Their security program becomes part of your risk surface.
The March 2023 data breach at OpenAI exposed chat histories of ChatGPT users. While the vulnerability was patched quickly, it demonstrated that AI providers face the same security challenges as any technology company.
Monitoring Limitations
Enterprise admin controls provide visibility into usage patterns. They don't provide content-level monitoring. You can see who used ChatGPT and how often. You can't automatically detect when someone submits a document containing SSNs or confidential client information.
DLP integration helps, but most organizations haven't fully integrated AI tools into their data loss prevention systems.
Prompt Injection and Manipulation
Enterprise documents processed by ChatGPT could contain malicious instructions attempting to manipulate model behavior. This attack vector exists regardless of enterprise tier.
Explainability
For regulated decisions (credit, employment, insurance), regulators increasingly expect explainnable AI. ChatGPT can't explain why it generated a specific response. This creates challenges for using AI outputs in regulated decision processes.
Enterprise Controls to Implement
Beyond subscribing to Enterprise, compliance requires additional controls:
Access Management
- SSO integration with corporate identity provider
- Role-based access limiting ChatGPT to approved users
- Multi-factor authentication required
- Regular access reviews removing departed employees
Usage Policies
- Acceptable use policy specific to AI tools
- Clear prohibitions on submitting specific data categories
- Guidance on reviewing AI outputs before reliance
- Requirements for human review of AI-assisted decisions
Data Classification Integration
- Clear mapping of data classifications to permitted AI uses
- Technical controls preventing upload of restricted data classifications
- Monitoring for policy violations
Audit Logging
- Enable comprehensive audit logging of AI interactions
- Retain logs per your data retention requirements
- Regular review of usage patterns for anomalies
- Integration with SIEM for security monitoring
Training
- Mandatory training before AI tool access
- Annual refresher on appropriate use
- Specific training for roles with access to sensitive data
- Documentation of training completion
Pre-Upload Safeguards
For documents containing sensitive information, implement redaction before AI processing:
- Automated identification of PII, PHI, and confidential information
- Redaction of identified sensitive data
- Audit trail of what was redacted
- Human review option for complex documents
This approach allows AI-assisted analysis while preventing sensitive data transmission.
Policy Framework Template
A compliance-oriented AI use policy should address:
Scope
- Which AI tools are approved
- Which roles have access
- Which use cases are permitted
Data Handling
- Data classifications prohibited from AI systems
- Required redaction for sensitive documents
- Approval requirements for specific data types
Usage Requirements
- Authentication and access requirements
- Logging and monitoring notice
- Output review requirements
- Prohibited use cases (automated decisions in regulated areas)
Incident Response
- Reporting requirements for suspected policy violations
- Response procedures for inadvertent sensitive data submission
- Escalation paths for AI-related security incidents
Governance
- Policy review frequency
- Exception approval process
- Compliance monitoring approach
Vendor Assessment Questions
When evaluating ChatGPT Enterprise (or renewing), ask:
Security Certifications
- Provide current SOC 2 Type 2 report
- Provide ISO certifications
- What is your penetration testing frequency?
- Provide evidence of vulnerability management program
Data Handling
- Confirm data is not used for model training
- What is data retention period for conversations?
- How is data deletion handled?
- What sub-processors access customer data?
Breach Response
- What is your breach notification timeline?
- Provide your incident response plan
- What breaches have occurred in the past 24 months?
- How were affected customers notified?
Regulatory Support
- Can you execute a DPA?
- Can you execute a BAA for healthcare use?
- What data residency options exist?
- How do you support data subject access requests?
Controls
- What admin controls exist for usage monitoring?
- What DLP integrations are available?
- Can we enforce SSO and MFA?
- What audit logging is available?
The Compliance Decision
ChatGPT Enterprise provides substantial security and compliance infrastructure compared to consumer versions:
- Independent security attestations
- Contractual data protection commitments
- No training on customer data
- Encryption and access controls
- Regional data residency
- Admin visibility and controls
These features make enterprise AI adoption feasible for regulated organizations. But feasible doesn't mean automatic. Enterprise features create the foundation. Compliance requires building appropriate policies, controls, and monitoring on that foundation.
The alternative is employees using consumer AI tools without any controls, which is what happens when organizations don't provide approved enterprise alternatives. From a risk perspective, controlled enterprise adoption beats uncontrolled shadow AI.
The compliance officer's role is ensuring that enterprise adoption comes with appropriate safeguards. The certifications and contractual frameworks help. The additional controls, policies, and ongoing monitoring complete the picture.
PaperVeil lets you redact sensitive information from documents before AI processing. Automated detection of PII, PHI, and confidential data. Audit trails for compliance documentation. The pre-upload safeguard that makes enterprise AI actually compliant.