Automatic
Reversible, bounded actions within existing authority can execute without additional intervention.
Governance
Capability does not create authority.
RSQS AI separates what a system can do from what it is permitted to do. Human authority, explicit release boundaries and evidence requirements are part of the technical architecture.
Control classes.
Automated actions are classified before execution so authority is explicit rather than inferred from technical access.
Review
Work may be generated or tested automatically, but acceptance is required before release or consequential use.
Authorise
The action cannot execute until a person with authority grants explicit approval.
Human in the loop
A human action is inherently required during execution, such as credential entry or an external validation step.
Prohibited
The action is blocked because the required authority does not exist or policy expressly disallows it.
Unknown / hold
Unknown is never silently converted into safe, lawful, true or authorised. The correct response can be to hold.
Hard boundaries
Automation must not manufacture legal authority, spending authority, public-release authority or irreversible consent.
Credential handling, material irreversible consequences, unresolved value judgements and public disclosure of protected intellectual property remain bounded by explicit human control. Prior approval is not treated as permanent approval for a different consequential act.
Private by default
Unclassified, uncertain, derived and internal material remains private by default. Owner secrets are never public. Proprietary core systems are not automatically promoted into public artifacts. Public interfaces expose only what is necessary for the approved purpose.