Know what it may use
Approved collections, currency, ownership and access are designed before the interface.
AI assistants / Governance
Put accountability around the assistant before giving it work.
Human review and AI governance define what an assistant may do, which information it may use, who approves outputs, how quality is evaluated and how problems are escalated. GA Applications turns those decisions into interface controls, permissions, logs, review queues and operating procedures suited to the particular use case rather than a generic policy document alone.
Bounded requestMay this draft proceed without a named reviewer?
RequestRetrieveDraftReview
Draft with evidence
The assistant has prepared a response, but it is not an approved decision.
1Approved use register / consequence tierapproved source
2Governance control / mandatory reviewcheck currency
Pending review. No external action is available in this demonstration.
Approved sourcesVisible evidenceRole permissionsHuman reviewEvaluationDisable path
Useful because the job is bounded. Trustworthy because the boundaries are visible.
AI assistance belongs inside a designed operating process. It does not replace accountable people, authoritative records or competent professional judgement.
Let people inspect support
Citations, source context and uncertainty travel with the draft.
Keep decisions attributable
Permissions, review, escalation, logging and shutdown routes match consequence.
01
Create a use-case record before development.
The record states audience, purpose, sources, output, decision consequence, model or provider, data categories and clear exclusions.
Value
Explain the user problem and why AI is considered.
Risk
Identify harm from wrong, missing, biased or exposed information.
02
Design review according to consequence.
A human gate must provide enough source context, time and authority to make a real decision rather than becoming a ceremonial click.
Reviewer
Name the competent role and alternatives when unavailable.
Interface
Show output, evidence, uncertainty, destination and effect together.
03
Define refusal, escalation and recovery.
The assistant needs visible routes for out-of-scope requests, source conflict, sensitive data, unsafe content and system failure.
Escalate
Send the user to an appropriate person or established process.
Disable
Allow authorised suspension without waiting for a code release.
04
Evaluate before launch and throughout operation.
Test sets, red-team scenarios, permission checks, feedback review and meaningful change control monitor whether the system remains within its approved purpose.
Baseline
Set acceptance criteria for support, citation, refusal and leakage.
Change
Re-evaluate model, prompt, source, permission and action changes.
05
Make ownership and evidence part of handover.
Documentation covers source owners, provider settings, access review, logs, retention, incidents, user guidance, monitoring and retirement.
Operations
Assign routine review and issue-response responsibilities.
Transparency
Tell users when AI is involved and how to challenge an outcome.
Questions worth answering
What teams ask before they commit.
Is human review always required?
The level depends on consequence and reliability. Consequential communication, system changes or decisions generally require meaningful review; limited low-risk assistance may use sampling or exception review after evidence supports it.
Does a governance policy make an AI system safe?
No. Governance must be implemented through product design, permissions, evaluation, operations and accountable decisions, then reviewed as the system changes.
Can we turn the assistant off quickly?
A defined disable or rollback path should be part of the design, with authority and user communication documented.
Govern before scaling
Map the purpose, evidence, human decision and stop path.
We can assess a proposed assistant and turn governance decisions into a testable product specification.