SŌVAI Proof of Concepts

SAAF — Sovereign AI Audit Framework

• Scope: methodology for sovereign AI auditing (processes, evidence, criteria).
• Outputs: evidence checklists, risk registers, compliance matrices.
• Compliance: EU AI Act (risk-based), GDPR (accountability), ISO/IEC 42001.
• Status: v0.2 internal; v0.3 public outline planned.
• Roadmap: integration with Audit OS metrics, annex templates (register, DPIA-lite).

SŌVAI Governance Matrix

• Purpose: mapping of responsibilities (business, data, model, ops, legal).
• Use: maturity scoring, gap analysis, remediation prioritisation.
• Results: domain views (security, safety, transparency, ethics).
• Status: v0.1 pilot; consolidation of labels & weighting ongoing.

SŌVAI Audit OS

• Goal: technical platform for evidence (datasets, models, metrics).
• Integrations: AIF360, Fairlearn (fairness), explainers (SHAP/LIME), traceability.
• Artifacts: signable reports, JSON attestations.
• Status: design + PoC modules; packaging ongoing.

SŌVAI COMPLY

• Goal: lightweight module for AI Act checks (categorisation, obligations).
• Functions: declarative rules (toml/yaml), completeness scoring, indicators.
• Status: CLI prototype → next: minimal web bundle.

SŌVAI Compliance Suite

• Goal: packaged “business” solution, designed as a sovereign and independent compliance platform.
• Target users: PMO / Compliance officers; Word/Excel import-export, decision traceability.
• Status: design phase; alpha after consolidation of SAAF + Audit OS.

3. Cross-cutting Modules

• SŌVAI-doc-RAG — local Retrieval-Augmented Generation on the SŌVAI knowledge base (privacy-by-default).
• AI Risk Taxonomy — enriched classification of AI risks (based on MIT AI Risk Repository, extended with multi-label & causal graph approaches).