Human-in-the-loop is often treated as a broad promise: a person will check the AI somewhere before the work is complete. That is not enough for legal operations.
An approval gate is a specific control
An approval gate defines who can authorise an action, what information they must see, which conditions trigger review and what the system records afterward. It turns oversight from a policy statement into an executable part of the workflow.
A low-risk internal summary may proceed automatically. Sending a notice, accepting a contractual position or communicating legal advice should route to a person with the appropriate authority.
Match control to consequence
Not every step requires the same friction. The strongest systems classify actions by consequence. Retrieval and organisation can often be automated. Recommendations may require review. External communication, binding commitments and material decisions usually require explicit approval.
Governance can improve speed
Controls do not necessarily slow work. A well-designed gate gives the reviewer the relevant source, proposed action, reason and exceptions in one place. The professional spends time on judgment instead of reconstructing context.
THAO, developed by Think Automata and backed by Alpha AI, separates reasoning, authorisation and execution so that organisations can configure these boundaries around their own policies and roles.