Agent safety

How much authority can this agent safely hold?

Set the smallest useful authority for each agent action, put human review before consequential effects, and rehearse a safe recovery path.

AI VISTA / DECISION01 / 03
WORKING DECISIONIncrease autonomy only after observation, approval, and recovery are already designed.
01INPUT
02CHECK
03BOUNDARY
RETURN TO THE REAL TASK✓

DECISION BRANCH

Answer the decision this topic actually owns.

  1. ASK FIRST↓

    What does each read, write, message, or payment change, and who can approve it before execution?

  2. IF↓

    Authority, review placement, and timeout recovery pass a drill: increase autonomous scope one action at a time.

  3. OTHERWISE✓

    Keep read-only or draft mode, name the human owner, and block unapproved external effects.

WORKED CASE

May a refund agent proceed all the way to payment?

The agent reads tickets, queries orders, calculates refunds, writes to CRM, and may call a payment tool. Read-only steps and balance-changing actions sit inside the same workflow.

WORKING ARTIFACTA permission ladder, human review gates, and a recovery runbook naming action owners, approval scope, unknown results, and rollback paths.
01WEAK SHORTCUT

Add one human confirmation at the final button without showing evidence or changes, and ignore duplicate requests and tool timeouts.

02BETTER JUDGMENT

Assign authority by action: reads may run, CRM drafts stay reversible, and payment requires one approval bound to the order and amount. Show evidence and effects before payment; name an operator to handle unknown results.

03ACCEPTANCE

Rehearse duplicate requests, payment timeouts, abnormal amounts, and rejected approvals. No duplicate refund occurs, partial success has an owner, and the operator can continue, reverse, or close from the record.

ENTRY & EXIT SIGNALS

Know when to enter—and when the decision is good enough to leave.

A topic is not an endless knowledge directory. Entry signals tell you whether the problem belongs at this layer. Exit signals decide whether to continue instead of substituting time spent reading for work completed.

AI VISTA / DECISION STATUSAgent safety
01ENTER HERE
  • An agent action may write, notify, or issue payment
  • Reviewers cannot see the evidence or approval scope
  • Repeated execution or partial success has no recovery owner
02LEAVE WHEN
  • Consequential actions have the smallest useful permission and exact approval scope
  • A reviewer can reject or change the action before execution
  • A timeout, repeat, or partial success has been rehearsed to a safe result

Once authority and recovery are explicit, compare task quality, end-to-end cost, and operating drift.

BOUNDARY

What this topic page will not do

The permission ladder guides control design; it is not organizational authorization or a security certification. Verify reversibility in the real system.

01Do not give an agent an administrator credential for every system02Do not use a vague confirmation to approve a specific payment or message03Do not raise authority without a human handoff and a recovery drill
When every action has an approver, review point, and recovery plan, carry those constraints into model and cost operations.