Tools / System Instructions

System Instructions

Versioned, downloadable system prompts that ground AI assistants in the CoSAI AI SRF. Enforce exactly one accountable party per activity, layer-cascading responsibilities, and autonomy classification for agentic systems.

Why this matters

AI assistants are increasingly used to draft governance documents, generate RACI matrices, and analyze deployment scenarios against regulatory requirements. Without a grounding system instruction, the outputs can contradict the SRF — misassigning accountability, skipping layer analysis, or treating "shared" responsibility as a valid final answer when the framework requires a single accountable party.

A canonical system instruction solves this by making the framework's core rules non-negotiable in any AI-assisted governance workflow.

What's planned

Core system instruction (v2.0) — A production-ready prompt that enforces the SRF's fundamental rules: one accountable party per activity, L1-to-L5 responsibility cascade, mandatory autonomy level classification (L0–L5) for agentic systems, and explicit citation of framework sections.

Prompt variants — Shorter, role-specific versions optimized for executive summary, auditor (evidence-focused), developer (technical controls), and legal/procurement (contract language) contexts.

Scenario packs — Few-shot examples for the 10–15 most common governance scenarios: third-party model evaluation, incident post-mortem, autonomy classification, contract clause generation, and sector-specific deployments in healthcare, finance, and public sector.

Version registry — All prompts are versioned and pinned to a specific SRF release. The registry includes a testing harness that validates outputs against framework rules.

Timeline

Phase 1 (0–3 months from v2.0 launch): core system instruction and initial variant set published to the GitHub repository.

← Back to tools