riskforge 0.1.2

EU AI Act Article 9 Risk Management System — open-source CLI

RiskForge

RiskForge is an open-source CLI that turns EU AI Act Article 9 compliance from a consultant invoice into a 30-minute developer workflow.

You answer 50+ guided questions about your AI system. RiskForge produces a SHA-256-signed Risk Management File (JSON + PDF) that satisfies Annex IV documentation requirements — ready for your legal team and your downstream compliance toolchain.

Built by AiExponent LLC. Apache 2.0. Runs entirely offline after pip install.

"The architecture is clean and the import boundary enforcement is the right call. The entry point plugin system will scale. The compound moat diagram is architecturally honest. Ship it." — Head of AI System Design, Google (design review, April 2026)

---

Quick Start

pip install riskforge

# 1. Register your AI system
riskforge init \
  --name "Loan Scoring Model" \
  --version "2.1" \
  --purpose "Automated credit scoring for retail loan applications." \
  --provider "Acme Financial Services" \
  --category essential_services

# Outputs:
# ✓ Initialised RiskForge project: Loan Scoring Model v2.1
#   System ID: f3a9c2d1-...

# 2. Run the guided 8-dimension risk assessment
riskforge assess f3a9c2d1 \
  --assessor-name "Alice Chen" \
  --assessor-role "AI Governance Lead"

# Walks you through 50+ questions across:
# health & safety, fundamental rights, discrimination, privacy,
# transparency, human oversight, robustness, data governance

# 3. Validate completeness before export (8 gates)
riskforge validate f3a9c2d1

# 4. Export your Article 9 Risk Management File
riskforge export f3a9c2d1 --format pdf --output loan-scoring-rmf.pdf
riskforge export f3a9c2d1 --format json --output loan-scoring-rmf.json

Result: A signed PDF your compliance team can file, and a rmf.json your downstream toolchain can consume — in under 30 minutes.

---

Why RiskForge

EU AI Act Article 9 requires providers of high-risk AI systems (Annex III: credit scoring, hiring, biometric identification, medical diagnosis, law enforcement, and more) to maintain a documented risk management system throughout the system's lifecycle.

The problem: No open-source tool existed for this. The alternatives were:

• Big 4 consulting: €80K–€350K per system
• Enterprise GRC platforms: $60K–$200K/year, designed for risk officers not engineers
• Spreadsheets: manually maintained, legally fragile, not verifiable

RiskForge is the engineering-native alternative.

---

What You Get

Output	Format	Use
Risk Management File	JSON (schema-validated)	Consumed by TransparencyDeck, ConformityBot, Sigil
Risk Management File	PDF (WeasyPrint, no system binaries)	Legal team review, regulator submission
Risk Management File	Markdown	Developer-readable summary, git-committable
Audit trail	JSONL (append-only, hash-chained)	Tamper-evident record for competent authorities

---

Features

Risk assessment

• 8 EU AI Act risk dimensions mapped to Article 9 obligations
• 50+ guided questions with per-question Article refs (Art.9, Art.10, Art.13, Art.14, Art.15) and NIST AI RMF / ISO 42001 cross-references
• Annex III pattern matching — pre-populates risk items for known use cases (credit scoring, hiring, facial recognition, etc.)
• 5×5 likelihood × severity scoring matrix with automatic band classification (low / medium / high / critical)
• Knowledge gap flagging — unknown answers generate test requirements via riskforge tests generate

Export and integrity

• SHA-256 hash-chained audit trail — every state mutation appended to audit.jsonl; any tampering detected by riskforge verify (exits code 2 for CI)
• JSON exports validated against rmf.schema.json (JSON Schema draft-2020-12) before every write
• PDF rendered by WeasyPrint — no LibreOffice, no wkhtmltopdf, no system binaries required
• Optional GPG/Sigstore signing with --sign-with key.pem

Validation

• 8 readiness gates that block export if the register is incomplete:
  • G1: all 8 dimensions covered
  • G2: Article 6(2) Annex III self-classification documented
  • G3: all high-risk items mitigated or accepted with rationale
  • G4: knowledge gaps have test requirements
  • G5: system metadata complete
  • G6: assessor identity recorded
  • G7: risk score distribution plausible
  • G8: no vague mitigation language detected

Integration

• Import accuracy evidence from rag-benchmarking: riskforge import --source rag-benchmarking --file report.json
• Import data lineage from TraceForge: riskforge import --source traceforge --file report.json
• Export consumed by TransparencyDeck (Art.13) and ConformityBot (Art.43)

Developer experience

• Zero outbound network calls after pip install (enforced by pytest-socket CI gate)
• State stored as YAML + JSONL — human-readable, git-diffable, PR-reviewable
• Runs on Linux, macOS, Windows (Python 3.11+)
• Plugin extensible — add question banks, exporters, adapters via pip; no config edit required

---

Architecture

RiskForge has four strictly-decoupled layers with CI-enforced import boundaries:

┌────────────────────────────────────────────────────────┐
│  CLI (Typer)                                           │
│  riskforge init / assess / validate / export / verify  │
│  Thin commands — no business logic                     │
└──────────────────────────┬─────────────────────────────┘
                           │ calls
┌──────────────────────────▼─────────────────────────────┐
│  Engine                                                │
│  AuditEngine · RiskEngine · ValidateEngine             │
│  AssessEngine · ExportEngine · TestDerivationEngine    │
│  No CLI or server imports (enforced in CI, ADR-02)     │
└──────────────────────────┬─────────────────────────────┘
                           │ reads/writes via
┌──────────────────────────▼─────────────────────────────┐
│  Storage (FileStore)                                   │
│  YAML + JSONL, chmod 600/700, async, pluggable ABC     │
└──────────────────────────┬─────────────────────────────┘
                           │ decoupled via adapter pattern
┌──────────────────────────▼─────────────────────────────┐
│  Integration Adapters                                  │
│  RAGBenchmarkingAdapter · TraceForgeAdapter            │
│  Discovered via Python entry_points (no hard imports)  │
└────────────────────────────────────────────────────────┘

Storage on disk:

your-project/
├── riskforge.yaml              # project manifest
├── .riskforge/
│   ├── audit.jsonl             # append-only hash-chained audit log
│   └── .nodelete               # sentinel
└── systems/
    └── <system-id>/
        ├── system.yaml
        ├── register.yaml
        ├── mitigations.yaml
        └── exports/
            └── rmf-*.json

State is plain YAML and JSONL — readable by regulators without RiskForge installed, reviewable in GitHub PRs, and merge-conflict-resolvable.

---

AiExponent Compound Moat

RiskForge is the structural centre of the AiExponent compliance toolchain. Every upstream tool produces evidence that feeds RiskForge; every downstream tool consumes its output.

[rag-benchmarking]  → benchmark_report.json → riskforge import → Robustness risks
[TraceForge]        → trace_report.json     → riskforge import → Data governance risks
                                                     ↓
                                              riskforge export
                                             /              \
                                    rmf.json                rmf.pdf
                                   /        \                   \
                    [TransparencyDeck]  [ConformityBot]   [Compliance Officer]
                       Art.13 docs        Art.43 cert

All connections are file-based. RiskForge never calls external APIs.

---

EU AI Act Article 9 Coverage

Article 9 Clause	How RiskForge covers it
9(1) — Establish and maintain RMS	Register lifecycle, version history, audit log
9(2)(a) — Identify known risks	Guided question bank, 8 dimensions
9(2)(b) — Estimate risks under misuse	Risk patterns, scenario-specific questions
9(2)(c) — Post-market monitoring integration	Import from VigilanceDash (roadmap)
9(2)(d) — Adopt risk management measures	Mitigation documentation with vague-detection
9(4) — Residual risk disclosure	Residual scoring + acceptance rationale export
9(7) — Testing requirements	riskforge tests generate — per-risk test requirements with metric hints
9(8) — Affected persons consultation	Human oversight dimension questions
9(9) — Vulnerable groups	Dedicated question bank entries, mandatory flag
9(10) — Documentation retention	Append-only audit log, export versioning
Annex IV — Technical documentation	Full RMF export structure

Cross-maps to: NIST AI RMF (GOVERN/MAP/MEASURE/MANAGE), ISO/IEC 42001 (Clauses 6.1, 8.4, A.6, A.7, A.9), Colorado AI Act SB 24-205, Texas HB 1709.

Disclaimer: RiskForge produces documented evidence for Article 9 compliance. It does not substitute for qualified legal counsel or notified body conformity assessment.

---

Contributing

The easiest contribution requires zero Python.

Add a question (no Python) — edit src/riskforge/_data/question_bank/<dimension>.yaml:

- id: HS-009
  text: "Your question here?"
  guidance: "Why this matters in one sentence."
  annex_iii_categories: [essential_services]
  default_likelihood_hint: 3
  default_severity_hint: 4
  article_refs: ["Art.9(2)(a)"]
  nist_rmf_ref: "MAP 1.5"
  iso42001_ref: "Clause 6.1"
  regulatory_status: settled

Add a risk pattern (no Python) — edit src/riskforge/_data/patterns/patterns.yaml.

Fix a bug or add a feature — see CONTRIBUTING.md for the full guide.

git clone https://github.com/aiexponenthq/riskforge
cd riskforge
make dev-setup   # pip install -e ".[dev]" + pre-commit install
make test        # 53 tests, all must pass
make lint        # ruff check + format

---

Privacy

RiskForge makes zero outbound network connections in CLI mode.

This is enforced in CI with pytest-socket --disable-socket. The --version output states it explicitly:

RiskForge v0.1.1 | Apache 2.0 | Zero telemetry | aiexponent.com

Your AI system's risk data never leaves your machine unless you explicitly deploy the optional API server (pip install riskforge[server]).

---

Releases

Version	Date	Highlights
v0.1.1	Apr 2026	Full riskforge assess implementation; PDF exporter fix; audit chain integrity fixes
v0.1.0	Apr 2026	Initial release — scaffold, models, engines, exporters, adapters

---

License

Apache 2.0 — free to use, modify, and distribute.

Built by AiExponent LLC — [e-mail]

---

Part of the AiExponent open-source AI governance toolchain: license-compliance-checker · rag-benchmarking · RiskForge