govern-ai-product-risk
Installation
SKILL.md
Govern AI Product Risk
Contract
Risk governance must be proportionate to real use and consequence. It does not replace legal, security, privacy, safety, or domain-expert review and must not claim compliance merely because a checklist is complete.
Workflow
- Define intended users, affected non-users, purpose, prohibited uses, deployment context, data, model and tool capabilities, autonomy, and decision consequences.
- Map harms across incorrect output, omission, bias, privacy, security, manipulation, overreliance, loss of control, unsafe action, economic impact, and downstream reuse.
- Rate severity, likelihood, exposure, detectability, reversibility, and distribution across relevant populations. Preserve uncertainty and credible worst cases.
- Select controls by layer: product scope, data, model, prompt, retrieval, tools, permissions, interface, human review, rate limits, logging, monitoring, and operations.
- Assign accountable owners, evidence requirements, approval rights, exceptions, expiry, and independent reviewers for high-impact decisions.
- Define misuse, abuse, red-team, accessibility, privacy, security, and domain tests with release-blocking thresholds.
- Establish monitoring, user reporting, containment, rollback, notification, remediation, and incident-review paths.
- Reassess when models, prompts, data, tools, autonomy, audience, geography, policy, or observed harm changes.