Policy & compliance

Compliance Bundles

IRIS ships compliance bundles that compile to Cedar rules and enforce at runtime — not just checklists.

Run a compliance check

iris compliance check --framework colorado-ai-act
iris compliance check --framework ccpa-admt
iris compliance check --framework china-pipl
iris compliance check --framework nyc-ll144
iris compliance check --framework illinois-ai-video
iris compliance check --framework aiuc-1
iris compliance check --agent my-agent --framework hipaa

Suggest applicable frameworks

iris framework suggest --agent my-agent

IRIS asks plain-English questions and recommends which frameworks apply and why — including California ADMT for employment/credit decisions, NYC LL 144 for ATS and hiring AI in New York City, Illinois AI Video for video interview analysis, China PIPL for users in China, and AIUC-1 for high-risk customer-facing agents (customer service, candidate scoring, interviewer agents).

Impact assessment & risk filing

iris compliance assess --agent my-agent

Generates a structured impact assessment stored in the Evidence Vault. Satisfies Colorado best practice (CO-002), CCPA ADMT risk assessment (CCPA-005), and PIPL PIPIA (PIPL-004). Export via iris evidence report for regulator submissions. See Evidence Vault for retention periods and the append-only data model.

Certify readiness

iris certify --agent my-agent --framework colorado-ai-act
iris certify --agent my-agent --framework china-pipl --format markdown
iris certify --agent my-agent --framework nyc-ll144 --format pdf
iris certify --agent my-agent --framework nist-ai-rmf --format markdown
iris certify --agent my-agent --framework aiuc-1 --format aiuc1-export
iris certify --agent my-agent --framework iso-42001

Produces a certification score with control pass/fail breakdown. For AIUC-1, --format aiuc1-export outputs JSON keyed by AIUC-1 sub-control IDs (B006.1, B006.2, etc.) for accredited auditors. For ISO 42001, IRIS shows a three-tier coverage breakdown (FULL / PARTIAL / NOT APPLICABLE) derived from AIUC-1's published crosswalk — not independent ISO interpretation. Pair with iris evidence report for auditor packages. Alias: iris test.

Free vs Pro bundles

Colorado bundles are free and ship with the open-source CLI. All other compliance bundles require an IRIS Pro license (iris license activate <key>). Free tier shows a 3-control preview for Pro frameworks.

Free tier

Bundle IDFrameworkStatus
colorado-ai-actColorado AI Act (SB 26-189)Available
colorado-chatbotColorado Chatbot Safety (HB 26-1263)Available
colorado-health-aiColorado AI in Health InsuranceAvailable
colorado-mental-health-aiColorado Mental Health AI (HB 26-1195)Available

Pro tier

Bundle IDFrameworkStatus
ccpa-admtCalifornia CCPA/CPRA ADMT RegulationsAvailable
china-piplChina Personal Information Protection Law (PIPL)Available
hipaaHIPAA for AI Agents (2025 NPRM + Section 1557)Available
soc2SOC 2 Trust Services Criteria (AI agent context)Available
gdprGDPRAvailable
nist-ai-rmfNIST AI Risk Management FrameworkAvailable
eu-ai-actEU Artificial Intelligence ActAvailable
fedrampFedRAMP ModerateAvailable
nyc-ll144NYC Local Law 144 — Automated Employment Decision ToolsAvailable
illinois-ai-videoIllinois AI Video Interview Act (820 ILCS 42)Available
aiuc-1AIUC-1 — AI Agent Security, Safety and ReliabilityAvailable
iso-42001ISO/IEC 42001:2023 (derived from AIUC-1 crosswalk)Available

SOC 2 accuracy note: The AICPA Trust Services Criteria do not include AI-specific controls. IRIS maps existing TSC controls (CC6, CC7, CC8, CC9.2) to AI agent governance evidence that auditors increasingly request. For AI-specific controls, use NIST AI RMF, AIUC-1, or ISO 42001 (IRIS's ISO mapping is derived from AIUC-1's published crosswalk).

AIUC-1 (voluntary certification standard)

AIUC-1 is a third-party certifiable standard for AI agents — Schellman is the first accredited auditor. IRIS does not issue AIUC-1 certificates; it generates technical evidence in AIUC-1's own evidence-ID format for controls IRIS enforces at runtime.

ControlObligationIRIS coverage
B006Prevent unauthorized AI agent actionstool_permissions + Cedar pre-execution hooks + Evidence Vault (FULL)
B007Enforce user access privilegesUser delegation + Evidence Vault audit trail (FULL)
C007Flag high-risk outputs for human reviewHITL gate + review records (FULL)
E015Log AI system activityEvidence Vault (FULL)
A006Prevent PII leakageDLP scanner (FULL)
E006Vendor due diligenceOUT OF SCOPE — legal/procurement process
C010–C012, D002, D004Third-party adversarial/hallucination testingOUT OF SCOPE — separate testing engagement
iris certify --agent my-agent --framework aiuc-1 --format aiuc1-export

ISO/IEC 42001:2023 (derived from AIUC-1)

IRIS's ISO 42001 coverage is composed from AIUC-1's official ISO crosswalk (dated 2025-09-18) plus IRIS's own AIUC-1 evidence map. IRIS does not independently interpret ISO 42001 clause language. Certification output shows FULL, PARTIAL, and NOT_APPLICABLE tiers — organizational clauses (leadership commitment, training, management review) are listed honestly, not hidden.

iris certify --agent my-agent --framework iso-42001

Run iris regulatory check --framework iso-42001 quarterly to verify the AIUC-1 crosswalk is still current.

California CCPA/CPRA ADMT (effective Jan 1, 2026)

CPPA finalized automated decision-making technology (ADMT) regulations in September 2025. Significant-decision requirements apply from January 1, 2027; risk assessment filing due April 1, 2028.

RuleObligationIRIS control
CCPA-001ADMT inventoryAgentPassport registration
CCPA-002Pre-use consumer noticepolicy-intent.md + app UI
CCPA-003Opt-out for significant decisionsuser_consent_logged + HITL
CCPA-005Privacy risk assessmentiris compliance assess
CCPA-0063-year record retentionEvidence Vault Pro (1095 days)

China PIPL (effective Nov 1, 2021)

Applies to any organization processing personal information of individuals in China — including US companies with Chinese users. Compliance Audit Management Measures effective May 1, 2025.

RuleObligationIRIS control
PIPL-001Lawful basis and explicit consentuser_consent_logged (blocks in prod)
PIPL-002Automated decision transparencypolicy-intent.md
PIPL-003Right to refuse automated decisionsHITL gate
PIPL-004Impact assessment (PIPIA)iris compliance assess
PIPL-005Cross-border transfer restrictionallowed_regions + RegionPolicy
PIPL-0063-year record retentionEvidence Vault Pro

HIPAA for AI Agents (2025 updates)

The January 2025 HHS NPRM proposes requiring AI tools in HIPAA risk analysis for the first time. Section 1557 (effective May 1, 2025) requires non-discrimination evaluation for AI in patient care decisions.

RuleObligationIRIS control
HIPAA-008AI tool inventory and risk analysisAgentPassport + iris certify
HIPAA-009AI non-discrimination in patient careiris red-team + compliance assess

IRIS does not generate BAAs — obtain one from your LLM provider before processing PHI in production.

NYC Local Law 144 — AEDTs (effective July 5, 2023)

The most actively enforced AI hiring law in the US. Applies to employers and employment agencies using Automated Employment Decision Tools (AEDTs) for candidates or promotions based in NYC. DCWP fines are $375–$1,500 per violation per day.

Free tier shows a 3-rule preview (bias audit, public disclosure, candidate notice). Pro unlocks accommodation routing (HITL) and 2-year audit retention.

RuleObligationIRIS control
LL144-001Annual independent bias auditbias_audit_date on passport + Evidence Vault; iris certify --framework nyc-ll144
LL144-002Public disclosure of audit resultsCertification PDF for legal review before publishing
LL144-003Candidate notice before AEDT usepolicy-intent.md + user_consent_logged
LL144-004Accommodation request mechanismHITL gate (Pro) — bypass AEDT for alternative review
LL144-0052-year audit documentation retentionEvidence Vault Pro (730 days)

Bias audit note: IRIS provides the technical governance documentation an independent auditor reviews. IRIS does not conduct the bias audit — the auditor must be a third party (e.g. Holistic AI, BABL AI, Pymetrics).

Illinois AI Video Interview Act (effective January 1, 2020)

Applies when AI analyzes video interviews for applicants for Illinois-based positions. Covers facial expression analysis, tone of voice, word choice, behavioral scoring, and any ML-based candidate scoring from video.

Free tier shows ILVI-001 through ILVI-003. Pro unlocks deletion workflow tracking and HITL human-review accommodation.

RuleObligationIRIS control
ILVI-001Applicant notification before AI analysispolicy-intent.md + app notification UI
ILVI-002Written consent before recordinguser_consent_logged with consent_type=ai_video_interviewblocks in production
ILVI-003Third-party sharing restrictionDeclared tool_permissions — undeclared vendors blocked
ILVI-004Delete recordings within 30 days of requestiris vault redact + deletion deadline tracking (Pro)
ILVI-005Human review on requestHITL gate for accommodation (Pro)

Consent is critical: ILVI-002 is the most frequently cited provision in enforcement. IRIS makes it technically impossible to run video analysis without logged consent in production.

Colorado AI Act — what IRIS covers

ObligationIRIS control
High-risk AI inventoryAgentPassport + iris scan
Impact assessment (CO-002)iris compliance assess
Transparency (CO-003)policy-intent.md
Consumer opt-out (CO-004)user_consent_logged context field
Annual review (CO-006)Evidence Vault

Evidence Vault and retention

Every compliance check, Cedar decision, and HITL action writes to the Evidence Vault as a signed, hash-chained event. Control mappings link each event to the frameworks it satisfies — one query powers Colorado, AIUC-1, ISO 42001, and HIPAA reports without re-collection.

iris evidence report --agent my-agent
iris evidence export --agent my-agent --format aiuc1 --output evidence.json
iris evidence record-cicd --system github_actions --run-id $GITHUB_RUN_ID --outcome success

Retention takes the longest period among attached controls (e.g. HIPAA 6 years beats Colorado 3 years). Full details: Evidence Vault guide.

Regulatory intelligence

iris regulatory check
iris regulatory list
iris regulatory watch # CI-friendly

See Regulatory Intelligence for full workflow.