Guardrails for Innovation: Making AI Safe for Public Sector Organizations