Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.complyhat.ai/llms.txt

Use this file to discover all available pages before exploring further.

ComplyHat ships four framework templates: SR 26-2, EU AI Act, NIST AI RMF, and ISO/IEC 42001. Each encodes which bias tests, protected classes, cadences, and report sections a regulator (or enterprise procurement) expects. When your host calls reports.start_draft for a framework, ComplyHat returns the template, structured evidence, and your compliance memory; the host composes the prose, the user approves, and reports.finalize persists the audit-tagged document. The wedge: EU AI Act Article 17. Every provider of a high-risk system on the EU market must keep a technical file ready for the regulator by 2 August 2026. ComplyHat’s EU AI Act template renders the Annex IV sections directly.

SR 26-2

Joint Fed/FDIC/OCC model risk management guidance (issued 2026-04-17). Documentation, validation, and ongoing monitoring for large bank holding companies.

EU AI Act

Regulation (EU) 2024/1689. Annex IV technical-file sections required by Article 17.

NIST AI RMF

NIST AI 100-1. Govern / Map / Measure / Manage attestations. Voluntary baseline, increasingly cited by examiners and named as a safe harbor by state AI laws.

ISO/IEC 42001

ISO/IEC 42001:2023 AI Management System. Annex A controls and management-review artefacts. The certification EU enterprise procurement increasingly requires.
Each template is versioned. Reports carry the framework version they were rendered against plus the trusted regulator URL and a host-agent directive, so before submission, your AI agent verifies amendments against the regulator’s published text and your counsel signs off. ComplyHat is a renderer, not a monitor; freshness verification is the host agent’s job, instructed by every MCP response.

Why these four

Two cover mandatory regulation in the largest AI markets (US banking + EU). Two cover the standard internal-methodology + external-certification pair enterprise buyers expect. State-by-state and vertical-specific frameworks (insurance, healthcare, employment) are tracked in the engine but not in the V1 customer surface; they may return as V2 verticals once the four-framework wedge is proven.

Capability matrix

Per-framework support for the five product capabilities.
FrameworkJurisdictionBias testingDrift monitoringModel cardsReportsOngoing monitoring
SR 26-2US Federal
EU AI ActEU
NIST AI RMFUS Federal
ISO/IEC 42001International
All four frameworks have a live report_templates row, so reports.start_draft returns a structured template for each, and the bias, drift, and explainability engines run framework-agnostically; what changes per framework is which obligations get rendered into the audit trail. Drift cadence defaults to quarterly and is configurable: SR 26-2 and the EU AI Act expect a fixed monitoring rhythm, while NIST AI RMF and ISO/IEC 42001 tie revalidation to your own model lifecycle and scheduled management review.

Bias-test coverage

FrameworkRequired testsCadenceUse case
SR 26-2Four-Fifths, statistical parityQuarterly (default)Model risk for large bank holding companies
EU AI ActFour-Fifths, statistical parity, equal opportunity, predictive parityQuarterly (default)Broader protected-class coverage for high-risk systems
NIST AI RMFFour-Fifths, statistical parity (mapped to Measure 2.11)Quarterly (default)Voluntary baseline / state-AI-law safe harbor
ISO/IEC 42001Four-Fifths, statistical parity (mapped to Annex A controls)Quarterly (default)AI Management System certification evidence
Every framework also requires a final model card. That is the extra check counted alongside the bias tests in each report’s evidence summary, so the EU AI Act report shows five checks (four bias tests plus the model card) while the other three show three.

Helpful resources

Authoritative regulator and standards body text first, then free explainers. Your counsel verifies the current text against the authoritative source; the explainers help your team get oriented. Every report is stamped with the framework version it was authored against, and frameworks.check_freshness returns the authoritative source URL so your host can confirm there have been no amendments before you submit.
SR 26-2 is supervisory guidance, not an enforceable rule, and it carves out generative and agentic AI. Treat it as the documentation bar examiners expect, not a statute.
For the underlying statistical methods, see methodology. For the MCP entry points, see the tool reference.
Last modified on May 30, 2026