Skip to content

Compliance Frameworks Reference

Arbitex ships 8 Compliance Bundles — pre-built Policy Packs with detection patterns and enforcement rules mapped to specific regulatory frameworks. This reference describes what each framework requires, how Arbitex controls address those requirements, and which Bundle to activate.


Each Compliance Bundle consists of:

  1. DLP detection patterns — Regex, NER, and contextual patterns targeting framework-specific data categories
  2. Policy rules — Enforcement actions (BLOCK, REDACT, CANCEL, ALLOW_WITH_OVERRIDE) mapped to rule conditions
  3. Audit references — Every enforcement action logs the specific framework requirement referenced (e.g., PCI-DSS Requirement 3.4)

Bundles activate through the Policy Engine and occupy a position in your policy chain. Multiple bundles can be active simultaneously for organizations with overlapping regulatory obligations.

Status Meaning
Active Control is in place in the current release
Planned Control is designed and on the roadmap; not yet available
Organizational Requirement that falls outside Arbitex scope; your organization is responsible

Bundle ID: pci_dss Standard: Payment Card Industry Data Security Standard v4.0 Scope: Any system that stores, processes, or transmits payment card data. An AI Gateway that proxies employee or customer conversations creates risk of cardholder data (PANs, CVVs) appearing in prompts.

The pci_dss bundle detects:

  • Primary Account Numbers (PANs) — all major card networks (Visa, Mastercard, Amex, Discover) with Luhn checksum validation
  • Card verification values (CVV/CVC) in proximity to PANs
  • Card expiration dates in card-context patterns
PCI-DSS Requirement Arbitex control Status
Req 3 — Protect stored cardholder data BLOCK action prevents PAN transmission through Gateway; matched category logged, content not retained Active
Req 3.4 — Render PAN unreadable BLOCK prevents PAN from reaching downstream model; audit log references PCI-DSS Requirement 3.4 Active
Req 6.4 — Protect web-facing applications Cloudflare WAF with OWASP ruleset + custom AI proxy rules; mTLS on all service connections Active
Req 8 — Identify users and authenticate API key authentication + RS256 JWT; SAML SSO; MFA enforced per org policy Active
Req 8.6 — MFA for all access MFA policy enforced at authentication layer; FIDO2/TOTP/SMS supported Active
Req 10 — Track and monitor all access Tamper-evident HMAC-chained audit trail; all API requests logged with org, user, action, timestamp Active
Req 10.3.2 — Protect audit logs 90-day active + 2-year archive; HMAC chain detects modification; SIEM export for long-term retention Active
Req 10.5 — Retain audit logs minimum 12 months 90-day Arbitex retention; SIEM integration required for full 12-month retention Active
BYOK / customer key management Bring Your Own Key (Epic H) Planned
Terminal window
POST https://api.arbitex.ai/api/v1/admin/orgs/{org_id}/compliance/bundles
Authorization: Bearer arb_live_your-api-key
Content-Type: application/json
{ "bundle_id": "pci_dss", "enabled": true }

Then add bundle_pci_dss to your policy chain. See Compliance Bundles for the two-step activation procedure.


Bundle ID: hipaa Standard: Health Insurance Portability and Accountability Act — Privacy Rule + Security Rule Scope: Covered entities (healthcare providers, health plans, clearinghouses) and business associates. AI Gateways used by healthcare organizations create risk of Protected Health Information (PHI) appearing in prompts.

The hipaa bundle detects all 18 HIPAA Safe Harbor identifiers:

  • Patient names in clinical context
  • Medical record numbers, health plan beneficiary numbers
  • Dates of service, birth dates, admission/discharge dates
  • Geographic subdivisions below state level (ZIP codes, street addresses)
  • Account numbers, certificate/license numbers
  • Device identifiers, URLs, IP addresses in clinical context
  • Biometric identifiers, full-face photographs
HIPAA Rule / CFR Reference Arbitex control Status
45 CFR § 164.312 — Technical safeguards BLOCK prevents PHI transmission; enforcement logged with framework reference Active
45 CFR § 164.502 — Uses and disclosures BLOCK prevents PHI disclosure to model providers or users outside permitted use Active
45 CFR § 164.312(b) — Audit controls Complete audit trail; PHI detections logged by category only, content not retained Active
45 CFR § 164.312(e)(1) — Transmission security TLS 1.2+ everywhere; mTLS on internal services and Outpost connections Active
Minimum Necessary standard Only matched category retained in audit, not PHI content itself Active
45 CFR § 164.314 — BAA requirements Business Associate Agreement available; contact account team Organizational
BYOK for PHI data encryption Bring Your Own Key (Epic H) Planned
Terminal window
{ "bundle_id": "hipaa", "enabled": true }

Bundle ID: gdpr Standard: EU General Data Protection Regulation (Regulation 2016/679) Scope: Processing of personal data of EU data subjects, regardless of where the processing organization is located. AI systems processing EU resident data through a Gateway are in scope.

The gdpr bundle detects EU personal data categories:

  • Names, email addresses, phone numbers, postal addresses
  • National identification numbers (EU member state formats)
  • EU financial identifiers (IBAN)
  • IP addresses (treated as personal data under GDPR)
  • Special category data: health indicators, religious/political affiliation patterns
GDPR Article Arbitex control Status
Art. 5(1)(c) — Data minimization Policy Engine enforces purpose limitation at AI layer; REDACT removes personal data before model sees it Active
Art. 5(1)(e) — Storage limitation Arbitex does not retain prompt/completion content; audit entries log category, not content Active
Art. 5(1)(f) — Integrity and confidentiality Encryption in transit (TLS 1.2+); HMAC-chained audit logs; BLOCK prevents unauthorized disclosure Active
Art. 25 — Data protection by design DLP pipeline runs before model invocation; policy controls enforced before data leaves tenant boundary Active
Art. 30 — Records of processing Audit trail with per-request logging; SIEM export provides records of processing activity Active
Art. 32 — Security of processing TLS everywhere; application-layer Fernet encryption; private endpoints; HMAC audit integrity Active
Art. 33 — Breach notification Audit trail provides detection evidence; incident response procedures are organizational Organizational
Art. 44–46 — International transfers Audit trail records cross-border data flows; SCCs and transfer assessments are organizational Organizational
BYOK / data sovereignty Bring Your Own Key (Epic H) Planned
Terminal window
{ "bundle_id": "gdpr", "enabled": true }

Bundle ID: glba Standard: Gramm-Leach-Bliley Act — Financial Privacy Rule + Safeguards Rule (16 CFR Part 314) Scope: Financial institutions — banks, lenders, insurance companies, investment advisors. The Safeguards Rule requires protecting customer nonpublic personal information (NPI).

The glba bundle detects financial NPI categories:

  • Account numbers and routing numbers
  • Social Security Numbers in financial context
  • Customer names combined with financial account context
  • Income, credit, and financial history indicators
GLBA Safeguards Rule Section Arbitex control Status
§314.4(c) — Risk assessment Audit trail provides AI system risk signal; detected NPI transmission events logged for risk review Active
§314.4(d) — Safeguards — access controls RBAC at API layer (admin/user roles); group-based model access controls; API key scoping Active
§314.4(d) — Safeguards — encryption TLS in transit; Fernet application-layer encryption for stored credentials and configs Active
§314.4(d) — Safeguards — NPI transmission prevention BLOCK action prevents NPI transmission through Gateway; detection patterns cover GLBA NPI categories Active
§314.4(h) — Monitor, test, respond SIEM integration exports GLBA-relevant enforcement events in OCSF format Active
§314.4(b) — Qualified individual Organizational governance; not an Arbitex control Organizational
§314.4(i) — Service provider oversight Arbitex functions as service provider; contract provisions are organizational Organizational
Terminal window
{ "bundle_id": "glba", "enabled": true }

Bundle ID: sox Standard: Sarbanes-Oxley Act — Sections 302 and 404 Scope: Public companies and subsidiaries. Section 404 requires assessment of internal controls over financial reporting (ICFR). AI systems used by finance teams are in scope when processing or generating financial reporting data.

The sox bundle detects financial reporting risk patterns:

  • Internal financial figures in draft reporting context
  • Material nonpublic information (MNPI) indicators
  • Earnings guidance and forecast language
  • Audit communication and financial committee patterns
SOX Control Area Arbitex control Status
ICFR — IT general controls: logical access API key auth + RS256 JWT + SAML SSO; RBAC enforced at middleware layer Active
ICFR — IT general controls: audit trail Tamper-evident HMAC-chained audit trail; all AI interactions logged with admin ID and timestamp Active
ICFR — Change management Audit log captures all policy chain changes with admin ID, timestamp, and before/after state Active
ICFR — Monitoring SIEM export of enforcement events; continuous monitoring of AI system behavior Active
Access controls for financial data Policy Engine restricts model access by group; group-based rules vary enforcement by user classification Active
Whistleblower protections Organizational governance; not an Arbitex control Organizational
Terminal window
{ "bundle_id": "sox", "enabled": true }

Bundle ID: bsa_aml Standard: Bank Secrecy Act / Anti-Money Laundering (31 U.S.C. § 5311 et seq.) Scope: Financial institutions, money services businesses, broker-dealers. AI systems used in customer communications or transaction processing are in scope for suspicious activity monitoring.

The bsa_aml bundle detects patterns associated with suspicious activity:

  • Transaction amounts and account identifiers in structured patterns
  • Currency transaction report (CTR) threshold indicators ($10,000+)
  • Structuring language and layering indicators
  • High-risk jurisdiction references
  • Correspondent banking and wire transfer identifiers
BSA/AML Requirement Arbitex control Status
Suspicious Activity Reporting (31 U.S.C. § 5318(g)) Suspicious patterns in AI interactions logged with matched category and rule reference; exported to SIEM for compliance team review Active
Customer Identification Program (CIP) Sensitive identification data in prompts triggers BLOCK or PROMPT action; detection patterns cover CIP-relevant identifiers Active
Recordkeeping (31 CFR § 1010.430) Audit trail retains enforcement records; SIEM retention should be configured for BSA minimum retention (5 years) Active
Currency Transaction Reports CTR threshold indicators trigger PROMPT governance for compliance team review Active
Compliance officer designation Organizational governance; not an Arbitex control Organizational
Terminal window
{ "bundle_id": "bsa_aml", "enabled": true }

Bundle ID: ccpa Standard: California Consumer Privacy Act (as amended by CPRA — Cal. Civ. Code § 1798.100 et seq.) Scope: For-profit businesses operating in California meeting revenue or data processing thresholds. AI systems processing California consumer personal information are in scope.

The ccpa bundle detects California consumer personal information categories:

  • Identifiers (name, email, IP address, account number, SSN)
  • Commercial information (purchasing records, financial history)
  • Biometric information
  • Geolocation data (precise location)
  • Professional and employment information
  • Sensitive personal information (SPI) under CPRA: government IDs, financial account credentials, health data, racial/ethnic origin, precise geolocation
CCPA / CPRA Provision Arbitex control Status
§ 1798.100 — Right to Know / Right to Delete Arbitex does not retain prompt/completion content; audit entries log category, not PI content Active
§ 1798.100(a)(3) — Data minimization (CPRA) Policy Engine enforces purpose limitation; groups with different authorized purposes have separate policy rules Active
§ 1798.150 — Security obligation Encryption in transit (TLS) and at rest (Azure platform AES-256 + Fernet application layer); audit trail for all PI processing Active
CPRA — Sensitive PI restrictions BLOCK action prevents unauthorized use or disclosure of CCPA sensitive PI categories Active
§ 1798.135 — Opt-out of sale/sharing Organizational signal; not an Arbitex control Organizational
Privacy policy and notice requirements Organizational governance Organizational
Terminal window
{ "bundle_id": "ccpa", "enabled": true }

Bundle ID: sec_reg_fd Standard: SEC Regulation Fair Disclosure (17 CFR Part 243) Scope: Issuers subject to Regulation FD. The rule governs selective disclosure of material non-public information (MNPI) — it is a disclosure rule, not a recordkeeping rule.

The sec_reg_fd bundle detects material non-public information and regulated financial records in AI traffic:

Entity type What it covers
insider_info Material non-public information
material_contract Material contract and trade agreement references
regulatory_action SEC regulatory enforcement records
bank_account_number Financial account identifiers
email Business communications
Reg FD consideration Arbitex control Status
Detection of MNPI before it leaves the organization DLP inline detection on the entity types above, across every configured channel module Active
Preventing MNPI disclosure to third-party AI providers Policy Engine BLOCK or REDACT action on bundle rule match Active
Evidence that a disclosure control was operating Tamper-evident audit trail; enforcement events exported to your SIEM via OCSF Active
Determining what constitutes “material” for your issuer Organizational judgement; not an Arbitex control Organizational
Public disclosure obligations after an inadvertent selective disclosure Organizational process; not an Arbitex control Organizational
Terminal window
{ "bundle_id": "sec_reg_fd", "enabled": true }

Organizations with overlapping obligations can activate multiple bundles simultaneously. Each occupies a position in the policy chain.

Terminal window
PUT https://api.arbitex.ai/api/v1/admin/orgs/{org_id}/policy/chain
Authorization: Bearer arb_live_your-api-key
Content-Type: application/json
{
"combining_algorithm": "first_applicable",
"packs": [
{ "pack_id": "bundle_hipaa", "sequence": 1 },
{ "pack_id": "bundle_pci_dss", "sequence": 2 },
{ "pack_id": "bundle_glba", "sequence": 3 },
{ "pack_id": "pack_01HZ_CUSTOM", "sequence": 10 }
]
}

A healthcare finance organization might activate hipaa, pci_dss, and glba simultaneously — all three bundles evaluate before custom organizational packs.


All enforcement events export to SIEM in OCSF v1.1 format. Each Compliance Bundle enforcement action maps to OCSF class 2001 — Security Finding.

Event type OCSF class
DLP enforcement (bundle match) 2001 — Security Finding
Authentication events 3002 — Authentication Activity
Administrative changes 3001 — Account Change

Filter SIEM dashboards using:

  • class_uid = 2001 — all enforcement events
  • cloud.account.uid = <org_id> — org-specific events
  • metadata.framework_reference — specific regulatory reference (e.g., PCI-DSS Requirement 3.4)

See Audit Data Model Reference for the full field-level OCSF mapping.