Compliance posture

What we've built —and what needs an auditor.

We don't claim compliance we haven't earned. This page lists every framework EvalGuard supports, and exactly what level of support means: control mappings, evidence engine, or third-party certified. Use it to evaluate fit honestly before you sign.

What the labels mean

Control mappings

We've linked each control in the framework to specific attack types, plugins, or platform features. Run a security scan; you get a per-control coverage report. Useful for auditor-pre-work + RFP responses.

Evidence engine

Every relevant platform action emits a tamper-detection-hashed evidence row mapped to the framework's controls. Auto-populated continuously from your audit log. Auditors consume the JSON directly via /api/v1/compliance/<framework>/attestation.

Certified

A third-party auditor has verified our controls through a full audit cycle. Requires the 6-month observation period (SOC 2 Type II), accredited certification body engagement, and YOUR own attestation against your specific deployment.

In progress

Foundation work shipped (data handling, encryption, policies) but the evidence engine + control mapping for this framework hasn't landed yet. Tracked in our 90-day roadmap.

Frameworks we support

OWASP LLM Top 10

Control mappings

All 10 controls mapped to attack plugins + scorers.

Every security scan returns a per-control coverage report you can show to your auditor or security team. Scan via /api/v1/security with attackTypes from the framework registry.

OWASP Agentic AI Top 10 (2025)

Control mappings

All 10 controls (AAI01-AAI10) mapped. First-class agent risks.

Includes plugins for memory poisoning (AAI01), tool misuse (AAI02), privilege compromise (AAI03), goal misalignment (AAI06), value-alignment drift (AAI07), repudiation/temporal-attack (AAI08), identity spoofing (AAI09), human-loop bypass (AAI10).

SOC 2 Common Criteria

Evidence engine

12 CC controls seeded; user-action + system-level evidence auto-collected.

Two evidence streams: (1) every audit-loggable user action (login, role change, key create/revoke, export, settings change, incident update) emits a SHA-256-hashed evidence row mapped to a SOC 2 control; (2) a background worker job collects synthetic system-level evidence hourly for controls that don't have a user action — CC6.1 membership review, CC6.6 api_key review, CC7.2 health-check, CC7.4 backup-run. Coverage dashboard at /dashboard/compliance/coverage. Auditor consumption: GET /api/v1/compliance/soc2/attestation (snapshot) + /api/v1/compliance/soc2/evidence (raw row export, paginated, payload_hash on every row). Type 1 attestation (point-in-time) target Q4 2026; Type II (6-month observation period + accredited firm) follows in 2027. Both gated on funding the auditor engagement — the evidence above is live regardless.

See live coverage →

EU AI Act (Regulation 2024/1689)

Evidence engine

Articles 9-15 mapped + Annex IV technical-documentation auto-generator.

High-risk system obligations: risk management (Art 9), data governance (Art 10), record keeping (Art 12), transparency (Art 13), human oversight (Art 14), accuracy/robustness/cybersecurity (Art 15). Annex III high-risk categories enumerated for deployer questionnaire flow. POST /api/v1/compliance/eu-ai-act/annex-iv auto-generates the 9-section Annex IV technical documentation from real platform state — audit logs, SOC 2 evidence, incident reports, applied standards — with SHA-256 tamper-detection hashing on the canonical JSON. Legal review of your specific deployment is your responsibility.

See live coverage →

ISO/IEC 42001:2023 (AIMS)

Evidence engine

Annex A controls A.2-A.10 mapped + Statement of Applicability auto-generator.

AI management system controls: policies (A.2), internal organization (A.3), resources (A.4), impact assessment (A.5), AI lifecycle (A.6), data (A.7), information for users (A.8), use (A.9), third-party relationships (A.10). POST /api/v1/compliance/iso-42001/statement-of-applicability auto-generates the SoA per Clause 6.1.3 — every Annex A control listed with applicability + implementation status derived from your real evidence collection. Coverage percentage computed across applicable controls. Accredited certification requires engagement with a recognized certification body.

See live coverage →

HIPAA

In progress

Data-handling implemented; third-party attestation Q2 2027.

Encryption-at-rest (Supabase AES-256), encryption-in-transit (TLS 1.3), org-isolated RLS, audit logging, BYOK provider keys via Supabase Vault. Healthcare vertical pack (medical + pharmacy plugins) ready. BAA-eligible only after attestation.

GDPR

Evidence engine

DSR intake, account-deletion atomic RPC, consent gates wired.

Data subject right intake at /api/v1/privacy/dsr. Right to erasure: account_deletion_full() Postgres function — 11-step purge in a single transaction. Consent enforcement at the gateway: HTTP 451 returned when subject withdraws. Data Processing Agreement template available; legal review required for your specific deployment.

NIST AI RMF

Control mappings

Govern, Map, Measure, Manage functions tied to platform features.

Risk Management Framework alignment via the existing compliance/nist-ai-rmf.ts registry. Coverage via security scans + drift detection + adversarial campaigns.

What this page is NOT

This page is not a substitute for legal review of your specific deployment. EU AI Act high-risk system obligations, HIPAA BAA requirements, and state-level AI laws (Colorado, California, etc.) all depend on how YOU deploy our platform. Use this page as evidence input to your own counsel; not as a green light for regulated workloads.