Appearance
GhostAI and Regulatory Frameworks: NIST, SOC 2, and Beyond
Executive Summary
GhostAI is designed to help organizations meet common regulatory and assurance requirements—including NIST, SOC 2, HIPAA, PCI-DSS, GDPR, and industry-specific frameworks—through data protection, auditability, access control, and flexible deployment options. This whitepaper is intended for security teams, compliance officers, and procurement evaluating AI platforms. It maps GhostAI capabilities to control areas and points to vertical-specific documentation for detail. GhostAI does not process or store sensitive content in its original form when sending data to AI models; detection, tokenization, and audit controls support compliance across these frameworks.
Frameworks in Scope
NIST
- NIST Cybersecurity Framework (CSF) and NIST SP 800-53 (e.g., access control, audit, encryption) are widely used in government and critical infrastructure. GhostAI provides controls that support compliance in these areas: access control and identity, audit logging, encryption of data and keys, and data residency options. The product itself is not certified against NIST SP 800-53. For environments that require formal NIST SP 800-53 validation (e.g., FedRAMP, CUI), private deployment and the UDS bundle support compliance validation via Lula/OSCAL; see deployment documentation.
SOC 2
- SOC 2 (AICPA Trust Services Criteria) addresses Security and, optionally, Availability and Confidentiality. GhostAI is offered in a context where SOC 2 Type II certification applies (inherited from the Dymium platform). Platform features map to common criteria: access control (RBAC, SSO, MFA), encryption (data and keys), audit controls (centralized and AI audit logs, configurable retention), and vendor risk mitigation (BYOK so inference can run in the customer’s environment).
Other Frameworks (Summary)
| Framework | Scope | GhostAI support |
|---|---|---|
| HIPAA | PHI privacy and security | BAA available; PHI detection and tokenization; audit controls; see Healthcare / HIPAA. |
| PCI-DSS | Cardholder data | Detection and tokenization of card data; access control; audit logging; see Financial Services. |
| SOX / GLBA | Financial reporting and consumer privacy | Access control, audit trail, data protection; financial vertical. |
| GDPR / CCPA | EU and California privacy | Compliance tagging; region selection (e.g., EU for GDPR); data handling and right-to-deletion support. |
| CUI / ITAR / FedRAMP / CMMC / NIST 800-171 | Government and defense | CUI marking detection, ITAR-aware protection, data residency; FedRAMP alignment for private/UDS deployment; see Government & Defense. |
| FTC Safeguards | Financial institutions (e.g., auto dealers) | Customer information protection, audit evidence; see Automotive. |
| FERPA | Student records | Education vertical; reconciliation and context controls; see Education. |
| ISO 27001 | Information security management | Controls align with common ISO 27001 control areas; position as alignment/readiness—certification is separate. |
How GhostAI Supports These Frameworks
Access Control and Identity
GhostAI implements role-based access control (RBAC), organization isolation, optional SAML SSO, and MFA support. Session management and automatic logoff options reduce exposure. These capabilities support SOC 2 CC6.x (logical access), NIST SP 800-53 AC family, and HIPAA administrative and technical safeguards for access control.
| Control area | GhostAI capability |
|---|---|
| Identity and access | RBAC (Viewer, Member, Admin, Owner; platform roles for multi-tenant); SSO (SAML 2.0); MFA |
| Isolation | Per-organization data and key isolation; mandatory per-tenant encryption keys (TEKs) for all orgs; organization data wipe scoped exclusively to the target org |
| Session | Session timeout, secure logout, browser-based (no local storage of sensitive data) |
Audit and Monitoring
Centralized audit logging records administrative actions and key operations. The AI Audit Log records every LLM call used for PII detection and verification, with configurable levels (summary vs. full) and floor/ceiling settings so compliance mandates (e.g., at least summary) can be enforced. Retention can support long-term retention (e.g., 7-year) for HIPAA and other requirements. Export and Compliance Reports support evidence for auditors. These support SOC 2 CC7.x (monitoring), NIST AU (audit), HIPAA audit controls, and PCI-DSS Requirement 10.
| Control area | GhostAI capability |
|---|---|
| Audit logging | Admin and key operations; AI pipeline events (detection/verification); configurable audit level (none / summary / full) |
| Retention | Configurable retention; 7-year retention option for HIPAA and similar |
| Evidence | Exportable logs; compliance reports; hash-based correlation without storing raw content |
Data Protection
Sensitive data is detected before it reaches any external AI provider. Over 64 entity types (PII, PHI, cardholder data, etc.) are identified via a multi-layer pipeline (Presidio, Dymium NER, vertical models, AI verification). Data is tokenized (replaced with opaque tokens) so the model never sees raw sensitive content; tokens are reversed only in the user’s session. BYOK keys are encrypted at rest with tenant-level encryption and are never logged. TLS 1.3 protects data in transit; optional end-to-end encryption is available in Enterprise. This supports SOC 2 Confidentiality, NIST SC/MP (system and communications protection, media protection), HIPAA technical safeguards, and PCI-DSS Requirements 3 and 4.
| Control area | GhostAI capability |
|---|---|
| Classification and handling | PII/PHI/PCI detection; vertical presets (healthcare, financial, government, manufacturing, etc.); per-vertical excluded entities to reduce false positives; SECRETS/DTSA compliance tagging; tokenization before external transmission |
| Encryption | Keys encrypted at rest with AES-256-GCM envelope encryption (per-tenant TEKs); TLS 1.3 in transit; optional E2E encryption (Enterprise) |
| Integrity | Checksums, token validation, no raw secrets in logs |
Vendor and Data Residency
Bring Your Own Key (BYOK) allows customers to use their own API keys for LLM providers so that inference can run in their cloud (e.g., AWS Bedrock BYOK) under their IAM and network controls. Private deployment allows GhostAI to run entirely in the customer’s environment so no sensitive data need leave their boundary. These options support data-residency and “your cloud, your keys” requirements common in HIPAA, PCI-DSS, SOC 2, and government contracts.
Incident and Risk
The audit trail supports incident investigation and breach notification readiness. Anomaly detection and access-pattern visibility are referenced in HIPAA/BAA context. The Compliance Score on the Admin Dashboard (0–100 with letter grade) reflects how fully the organization has configured security and compliance features (SSO, data residency, multiple admins, PII protection, audit logging) and provides a readiness indicator; grade A is described as meeting enterprise compliance requirements. Use the dashboard and Compliance Reports for ongoing monitoring and evidence.
Deployment and Scope
| Model | Best for |
|---|---|
| SaaS | Standard compliance needs; customer relies on vendor isolation, encryption, and logging; BAA and contract terms apply. BYOK (e.g., Bedrock) keeps inference in the customer’s AWS. |
| Private | Strict infosec or regulatory environments (e.g., NIST/FedRAMP/CUI). GhostAI runs in the customer’s environment; no data need leave their boundary; IAM-only options avoid storing provider keys. |
| UDS / on-prem | Government and defense; NIST SP 800-53 validation support via Lula/OSCAL in the UDS bundle. |
For the strongest data-residency and control story, private deployment with BYOK (or IAM-only where applicable) is recommended. See AWS Bedrock BYOK and Deployment documentation.
Operational Guidance
- Select the right vertical — In Settings → Industry Vertical, choose Healthcare, Financial, Government, etc., to enable relevant detection and tagging.
- Set audit level — Use audit floor/ceiling (e.g., floor
summary) to meet compliance mandates; see AI Audit Logs. - Use Compliance Score and reports — Review the Admin Dashboard Compliance Score and Compliance Reports; act on recommendations (SSO, data residency, multiple admins).
- HIPAA — Execute a BAA where required; use Healthcare Shield add-on for HIPAA-grade PHI detection and reporting.
- Data residency — Configure region (and BYOK where applicable) so inference and data stay in the required jurisdiction.
- Periodic review — Conduct access reviews and audit log review on a schedule appropriate to your framework.
References
Internal
- HIPAA Compliance
- Financial Services / PCI-DSS
- CUI & ITAR
- BYOK
- AI Audit Logs
- Compliance Reports
- Admin Dashboard & Compliance Score
- AWS Bedrock BYOK
- Knowledge Reconciliation & Data Leakage Prevention
External
Voice Biometric and Audio Privacy
Overview
GhostAI's Voice Assistant feature introduces audio data into the privacy pipeline. Voice data carries unique regulatory obligations because voiceprints are classified as biometric identifiers under several frameworks.
Regulatory Mapping
| Framework | Voice Requirement | GhostAI Control |
|---|---|---|
| BIPA (Illinois) | Written consent before collecting biometric identifiers (including voiceprints); retention schedule and destruction policy | Explicit consent flow before voice activation; configurable retention; no voiceprint storage by default |
| GDPR Article 9 | Biometric data is special category; requires explicit consent or other Article 9(2) basis | Consent captured with timestamp and purpose; local-first STT avoids cross-border audio transfer |
| HIPAA | Voice recordings containing PHI are ePHI | Voice PII Shield vocalizes category labels instead of PHI values; local STT keeps audio on-server |
| CCPA/CPRA | Biometric information is personal information; right to delete | No persistent audio storage; consent withdrawal disables voice features |
| FERPA | Student voice recordings may be education records | Voice features respect vertical configuration; education vertical defaults to local-only processing |
Privacy Architecture
GhostAI's Voice PII Shield ensures that audio data receives equivalent or stronger protection than text:
- Outbound (TTS): Text is vocalized through
TokenVault.vocalize()before reaching any TTS provider — PII values are replaced with category labels (e.g., "a social security number" instead of the actual SSN) - Inbound (STT): Local-first speech-to-text (faster-whisper) processes audio on-server by default; cloud STT is opt-in and configurable per privacy level
- No voiceprint storage: GhostAI does not create, store, or transmit voiceprint templates
- Consent management: Voice activation requires explicit user consent with BIPA-compliant disclosure; consent status is audit-logged
Privacy Levels
| Level | TTS Provider | STT Provider | PII in Audio | Recommended For |
|---|---|---|---|---|
| Maximum | Local only | Local only | Category labels | HIPAA, BIPA, government |
| Standard | Cloud (vocalized) | Local-first | Category labels | General enterprise |
| Relaxed | Cloud (full PII) | Any | Full PII | Low-sensitivity use cases |
Operational Guidance
- Select privacy level — In Settings → Voice, choose the privacy level appropriate to your compliance posture
- Review consent flow — Ensure the voice consent disclosure meets your jurisdiction's requirements (BIPA requires specific statutory language)
- Audit voice events — Voice consent grants and revocations are logged in the AI Audit Log
- Data residency — For Maximum privacy level, all audio processing occurs on-server with no external transmission
Disclaimer
This whitepaper does not constitute legal or certification advice. Organizations are responsible for validating their compliance with applicable frameworks and for working with their own auditors and legal counsel. GhostAI provides controls and documentation to support that process; certification or attestation status (e.g., SOC 2 Type II) applies in the context described by the vendor and may be subject to scope and timing.