EVIDENCE & LIMITS
Original evidence: the human-approval rules used while building Tim Yslava’s website, analytics ownership, content controls, and distribution system during the week ending July 30, 2026. NIST’s AI Risk Management Framework provides an independent public reference for defined human roles and oversight.
The short answer
AI can draft, sort, flag, and record work. A person should make the call when a mistake could hurt money, people, privacy, trust, or work that is hard to undo.
This does not mean a person must click every small button. It means the team must name the choices that carry real weight and give one person the duty to approve them.
This rule came from real work
We used this rule while setting up Tim’s website, analytics, content checks, and social accounts. AI prepared steps, checked pages, and kept a record. Tim handled sign-ins and owner approval.
When a Meta business account showed a restriction, we stopped asset changes. When social routes were not ready, we paused posting but kept the safe website work moving. When a page went live, we checked the public link before calling it done.
Keep money choices with a person
AI can sort bills, compare choices, or prepare a report. It should not move money, set a binding price, place a trade, or accept a deal unless an authorized person makes that choice.
The person needs the amount, reason, limit, and likely downside. If those facts are missing, the work is not ready for approval.
Keep people and private facts protected
A model may miss history that matters to a worker, customer, partner, or family member. Use it to prepare facts, not to make the last call on a choice that can change someone’s work or standing.
Treat private facts the same way. Give the tool only what the job needs. Remove contact, customer, login, money, health, and family details that do not belong in the task.
Check every public claim
A fast draft can still put the wrong fact on the wrong account. Before a public post, page, or reply goes live, check the source, name, date, account, audience, and next step.
Then check the result. A saved file is not a live page. A row in a plan is not a scheduled post. Keep the public link or platform record.
Put a gate before hard-to-undo work
Deleting an account, moving ownership, signing a deal, changing access, or locking a system may be hard to fix. AI can show the steps and prepare a preview. A person should approve the exact action.
Write the safe way back before the change. If there is no clear owner or no safe way back, stop and choose a smaller step.
Use three questions on any new task
Ask three things: Who could be hurt if this is wrong? Can we undo it? Who owns the result? These questions are quick enough for a small business and strong enough to catch many bad ideas.
NIST’s AI Risk Management Framework uses a more formal version of the same idea. It calls for clear human roles, clear oversight, and a written record. Start with the plain checklist here, then make the controls fit the size and risk of your work.
TRY THIS
The five-part human check
Ask who could be hurt, who owns the choice, and whether the work can be undone. Use a person for each choice below.
- 01 · MoneyCan this move, spend, price, invest, or risk money?Let AI prepare the facts. A named person makes the choice.
- 02 · PeopleCan this affect a job, customer, partner, or family?A person checks the facts, context, and likely harm.
- 03 · Private factsCan this expose data that should stay private?Stop, remove what is not needed, and ask the data owner.
- 04 · Public trustWill this publish a claim, message, or decision?A person checks the truth, account, audience, and final copy.
- 05 · Hard to undoCan this delete, transfer, sign, or lock something?Require a preview, clear approval, and a safe way back.
SOURCES
Check the original sources
What to remember
- Keep people in charge of money, people, privacy, public claims, and hard-to-undo work.
- Name one person who owns each important choice.
- Give AI only the facts it needs and remove private details.
- Check the real result and keep a safe way back.
STAY IN TOUCH
Follow the record as it is built.
Tim shares plain-English lessons from building companies, serving communities, and improving institutions—work that is developing the judgment and operating range for broader responsibility over time.