Part 4: What to Watch Out For

Is it OK to put this into AI?

Starting today, the story changes a bit. So far it was "What It Can Do"; from here on it's "What to Watch Out For." And the first answer is this — put things in assuming that once you type it in, it's out of your hands.

Leave out things that can pinpoint you, like your name, address, phone number or passwords, and then ask. Keep only what you're asking about, and the answer comes out the same.

Be even more careful with info that isn't yours. Where your friend's face or contact info ends up isn't your call to make.

Try it — Is it OK to put this into AI?

Here are six situations. Before you put something in, think just once: is it OK if this stays in someone else's hands?

Pieces point to you when they're combined

Each piece on its own says nothing, but stick a few together and it narrows down to one person.

Privacy laws can work this way too — South Korea's, for example, counts information as personal if it can easily be combined with other information to tell who someone is.

1. Q1What's the clearest example of something you should never put into AI?

   1. ① A photo of a math problem
   2. ② The title of a book you read
   3. ③ Your favorite color
   4. ④ Your password

2. Q2When do small pieces of information become dangerous?

   1. ① When several pieces get combined
   2. ② When there are lots of words
   3. ③ When you ask at night
   4. ④ When you ask in another language

3. Q3Uploading a photo of your friend's face?

   1. ① Fine, it's my info
   2. ② No, it's their info
   3. ③ Photos don't matter
   4. ④ Just remove the background

Once you type it in, it's out of your hands, and pieces point to you when they're combined.

Why did the meaning of "copying" get shaky?

Before AI, "copying" was judged by how much the result looked like the original. You just put the two side by side.

After AI, one more question got added — is it OK that it learned from other people's work before making anything? Now there are fights even when the results don't look alike.

This still has different answers in different countries, and big court cases are still going on. So don't memorize "this is illegal" — it's better to know what the issue actually is.

Try it — What changed before and after AI?

Switch the same question between "Before AI" and "After AI." The rules didn't disappear — there's just one more thing to judge.

Judging only the result vs judging what it learned from too

The rules didn't go away — there's just one more box to judge.

South Korea's government guide even notes that until more court rulings pile up, the call stays unclear. Not knowing is normal right now.

1. Q1Before AI, how was "copying" judged?

   1. ① How much the results looked alike
   2. ② Who was more famous
   3. ③ Who posted first
   4. ④ How expensive it was

2. Q2What question got added after AI?

   1. ① How many pictures it drew
   2. ② What it learned from before making it
   3. ③ How many colors it used
   4. ④ How many seconds it took

3. Q3Where do fights over training data stand now?

   1. ① Ruled all illegal
   2. ② Ruled all legal
   3. ③ Varies by country; ongoing
   4. ④ Nobody's fighting

We used to look only at the result, but now we ask what it learned from, too.

Is this video real?

Honestly? You can't tell just by looking. The old trick of counting fingers barely works anymore.

So change where you look. Instead of picking apart what's on the screen, follow where it came from — is it where it should be, and did anyone else film it too?

The people making this stuff are starting to label it too. Europe has put rules into law for planting invisible marks and for saying when AI made something, and other countries are working on their own.

Try it — Trust it or doubt it?

Six situations. Don't judge how well it's made — read the clues that come with it.

Clues that stopped working vs clues that work now

Use the tricks you learned a few years ago and you'll actually get fooled.

Tech that plants invisible marks is in use too, and Europe has rules for labeling AI-made content and making deepfakes say what they are.

1. Q1These days, what's the more reliable way to spot a deepfake?

   1. ① Count the fingers
   2. ② Check whether the picture quality is good
   3. ③ Trace back where it came from
   4. ④ See whether it's short

2. Q2Your friend's voice asks you to send money, fast. What do you do?

   1. ① It's their voice, so send it
   2. ② Post it in the group chat first
   3. ③ Send just a little
   4. ④ Check it's really them some other way first

3. Q3A video that says it was made with AI is...

   1. ① Not a trick if it says so
   2. ② Always a deepfake
   3. ③ Always illegal
   4. ④ Not allowed to be watched

Stare at the screen all you like, you won't see it — follow where it came from.

Is someone missing from this answer?

AI doesn't decide to dislike anyone. If the data it learned from leans one way, its answers lean that way too — that's all.

In 2018, a study tested programs that guess from a face photo whether it's a man or a woman. They got it wrong most often for darker-skinned women and least often for lighter-skinned men.

So there's no need to be scared — just build one habit. When a question sorts or picks people, ask once more: 'What about other cases?'

Try it — Is there bias in this answer?

Read each situation and pick one of the two. You'll get a feel for which questions tend to lean and which ones don't.

How a lean flows through

Bias doesn't suddenly pop up in one place — it flows in from the start.

Fixing the AI is hard, but asking once more at the very last step is something you can do today.

1. Q1What's the biggest reason AI ends up biased?

   1. ① AI dislikes people
   2. ② The internet is slow
   3. ③ It's bad at math
   4. ④ The data it learned from leans one way

2. Q2What did the 2018 face-photo study show?

   1. ① Most mistakes on darker-skinned women's faces
   2. ② The same mistakes on every face
   3. ③ Mistakes only on blurry photos
   4. ④ Always more accurate than people

3. Q3Which question is most likely to get a biased answer?

   1. ① What's 3 times 7?
   2. ② Who'd make a good class rep?
   3. ③ Water's boiling point?
   4. ④ How do I change the wifi?

AI doesn't hate anyone. It just leans toward what it saw most. When a question picks people, ask once more: 'What about other cases?'

Where did I stop and AI start?

Saying you used AI isn't confessing to something. It's a label that shows which part is your own work — if you don't say it, even the parts you did well get doubted.

Doing it is easier than you think. Split the assignment into five pieces — topic, sources, draft, polish, fact-check — and write one line on who did each. That's it.

Pictures and videos sometimes get invisible marks too. Google puts invisible watermarks in what its models make, and the European Union has written into law a rule that AI-made content must carry a mark machines can read (when it kicks in is still being adjusted).

Try it — Split the assignment into pieces and label them

Compare the same book report done three different ways. Orange is a piece AI did, green is a piece you did, and faded means nobody did it.

If you don't say / If you say

Same assignment — but writing that one line or not makes this much difference.

Write just three things — which piece, how you used AI, and what you changed — and your one line is done.

1. Q1What's the biggest reason to say you used AI?

   1. ① To clearly protect the part I did
   2. ② To get in less trouble
   3. ③ Because the AI asks you to
   4. ④ To hit the word count

2. Q2What easy way to disclose AI use did you learn today?

   1. ① Just write 'used AI' at the top
   2. ② Split the task into pieces and note each one
   3. ③ Change the text color
   4. ④ Make it shorter

3. Q3What's an invisible mark put into a picture called?

   1. ① A bookmark
   2. ② A password
   3. ③ A watermark
   4. ④ A screenshot

Saying so isn't confessing — it's a name tag. One line on which piece you got help with is enough.

If AI does it all, what's left for you?

The test isn't a long rulebook — it's one sentence. When you're done, did something stay in your head?

Hand the whole thing over and all you have is the finished product — you can't explain what's in it. That shows right away when you're presenting or sitting in front of a test.

But ask only about the spots where you're stuck and rewrite it yourself, and you keep both the work and the skill. Same tool — what changes is where you use it.

Try it — Handing it all over vs doing it together

Imagine doing the same homework two ways. Switch back and forth between them and compare what's left when you're done.

Four steps when you face homework

Just keep this order and you won't slide into handing it all over.

If you can't get past the fourth box, it isn't your homework yet — just go back one box.

1. Q1What decides whether it's OK to use AI on homework?

   1. ① Whether you get caught
   2. ② Whether something stays with you at the end
   3. ③ How much time it saves
   4. ④ Whether you hit the word count

2. Q2What's the biggest problem with handing all your math homework to AI?

   1. ① The blanks don't get filled
   2. ② Your handwriting gets neater
   3. ③ You get stuck again on the same type
   4. ④ It takes longer

3. Q3What's a better question when you're stuck?

   1. ① Just give me the answer
   2. ② Hand it in for me
   3. ③ Rewrite all of it
   4. ④ Show me where I got stuck

What you hand to AI is the stuck spot, not the whole homework. If you can explain it in your own words at the end, you did it.

Can you trust this story?

Spotting fake news isn't about having a sharp eye — it's about order. Check three things in turn and most of it gets filtered out.

Who said it first (source), when it happened (date), and whether it shows up anywhere else (cross-check) — if even one of these is empty, stop.

Now that AI can make text and photos in seconds, 'it looks convincing' doesn't count as evidence anymore. Don't trust your eyes — trust the clues.

Try it — Can you trust this story?

These are situations you could really run into in a group chat or your feed. Keep the three clues in mind — source, date, cross-check — as you pick.

Three boxes before you believe it

Before you share something, just pass through these three boxes. All of it takes less than a minute.

The strongest one is the third box — checking whether other places reported it separately and say the same thing.

1. Q1What are the three clues to check before you believe a story?

   1. ① Length, font, number of photos
   2. ② Headline, emoji, tone
   3. ③ Views, comments, likes
   4. ④ Source, date, cross-check

2. Q2Because of AI, what can't you use as evidence anymore?

   1. ① It looks convincing
   2. ② There's a link to the original
   3. ③ It has a date on it
   4. ④ Several places ran it

3. Q3Which of these is most trustworthy?

   1. ① Group-chat post, no source
   2. ② Notice with original link
   3. ③ Unnamed expert's claim
   4. ④ Photo from 3 years ago

Looking convincing isn't evidence anymore. Source, date, somewhere else — judge by those three.

What if a call comes in your family's voice?

AI scams always follow the same steps. Copy a voice or face → pretend to be someone you know → rush you → ask for money or information.

The FBI in the US has put out an official warning that AI-made voices, photos and videos are being used in scams. Even a short voice clip is enough to copy someone pretty closely.

Stopping it is almost boringly simple. Hang up, then call back yourself on the number you already knew. If it really is your family, calling back changes nothing.

Try it — The order of a scam, and the order for cutting it off

Tap the mixed-up pieces in order. Once you know the path a scam takes, you can see where to cut it off.

What scams aim for / What breaks them

Scams aim for three things. Flip each of them, one at a time.

The more urgent it feels, the more a pause is the fastest way through — something truly urgent will still be urgent 5 minutes later.

1. Q1What's the surest way to stop an AI scam call?

   1. ① Hang up and call back on a number you know
   2. ② Tell the real voice apart by ear
   3. ③ Stay on the call longer
   4. ④ Ask the caller their name

2. Q2Why do scammers push 'it's urgent'?

   1. ① They really are out of time
   2. ② So you don't get time to check
   3. ③ To save on their phone bill
   4. ④ To be polite

3. Q3A friend asks for an urgent favor from an unfamiliar account. You...

   1. ① Send it right away
   2. ② Ask that account back
   3. ③ Check via your usual contact
   4. ④ Tap the link to see

A voice can be copied, but a call you make yourself can't.

How much should you hand over?

This isn't about not using it. It's about whether you're the one deciding how much you hand over — that's all.

In 2025, a research team that ran a writing experiment reported that the people who handed more to AI had trouble remembering sentences they had just written, even though the writing itself was fine. It's a preliminary study still under review, so don't memorize it as the final word.

So the order matters more than the share. Give it 5 minutes with your own brain first, then hand it over — same tool, but what sticks with you changes.

Try it — Turn up how much you hand over

Slide the handle from 0 to 100 and watch how three things move. The bar numbers aren't real measurements — they're drawn to show the trend.

How much you hand over vs what sticks

Not 0, not 100 — the spot where the most sticks with you is the middle.

The middle bar is longest for one reason only — it's the only way that leaves you with both the work and the skill.

1. Q1What does today's 'not leaning on it too much' mean?

   1. ① Never using AI at all
   2. ② Deciding for yourself how much to hand over
   3. ③ Using it only once a day
   4. ④ Handing over only the hard stuff

2. Q2In the 2025 writing experiment, the people who handed more to AI...

   1. ① Wrote worse essays
   2. ② Took more time
   3. ③ Had trouble remembering their own sentences
   4. ④ Made spelling mistakes

3. Q3How do you get more to stick, using the same tool?

   1. ① Ask over and over
   2. ② Hand it over from the start
   3. ③ Do it alone to the end
   4. ④ Try 5 min first, then ask

Don't hand it all over, and don't skip it entirely. Give it 5 minutes with your own brain first — just changing the order changes what sticks.

Does every AI follow the same rules?

AI has laws too. But they don't treat every AI the same way — they sort uses into boxes by how risky they are and set different rules for each.

The European Union's AI Act shows this most clearly. It splits uses into four boxes: banned outright · high-risk · must disclose · minimal.

In South Korea, the AI Basic Act also took effect in January 2026. One rule that stands out: content made by AI has to be labeled as made by AI.

Try it — Same list, different boxes

Here are seven ways AI gets used. Tap each risk box: only the uses in that box light up, and the rest fade.

The bigger the risk, the stricter the rules

Even the same AI gets rules of different strength depending on what it's used for.

The four-box framework the EU AI Act uses — South Korea's AI Basic Act also singles out high-risk uses for tougher treatment.

1. Q1Which is closest to how AI laws set their rules?

   1. ① The same rules for every AI
   2. ② Ban all new AI
   3. ③ Different rules for different levels of risk
   4. ④ Let the makers decide

2. Q2What happens to a use in the 'high-risk' box?

   1. ① It can't be used at all
   2. ② There are no rules
   3. ③ Just change its name
   4. ④ It must pass checks and run with human oversight

3. Q3What rule applies to an AI-made video?

   1. ① Say AI made it
   2. ② It's banned
   3. ③ Lower its quality
   4. ④ Make it shorter

The law doesn't block AI completely, and it doesn't leave it alone either. It sorts uses into boxes by risk and treats each one differently.