A Family Code Word. The One Practice That Handles the Scariest AI Risk of All.

This is our reading of Bill Gates's 2023 essay The risks of AI are real but manageable (Gates Notes). Read the original →

Read the paragraph Gates wrote in 2023.

"If you get a voice message that sounds like your child saying 'I've been kidnapped, please send $1,000 to this bank account within the next 10 minutes, and don't call the police,' it's going to have a horrific emotional impact far beyond the effect of an email that says the same thing."

This is not science fiction. This is the scene he warned about three years ago. Today the FBI and police departments in China are both handling a rising number of "AI voice-clone extortion" cases. All a scammer needs is 30 seconds of public audio of your child. One short video, one TikTok, one Douyin clip is enough. Out of that comes a voice that sounds exactly like your child in a panic. Perfectly, chillingly, real.

You might read this and think: how am I supposed to defend against this? Can it even be defended against?

Gates gives two reasons not to despair.

The first is historical. He wrote his sentence in 2023: "The best reason to believe that we can manage the risks is that we have done it before." Remember the Nigerian prince scam? Millions fell for it at first. Today you see one and barely stop reading. People can learn. It takes time. For deepfakes, we simply have to learn faster.

The second reason: there is one thing you can do tonight at the dinner table that works almost every time.

It is called a family code word.

Sit down with your kids and your partner and agree on a word only your family knows. It can be the name of your first dog, a dish at a hole-in-the-wall restaurant you used to go to, a random tiny event from an old vacation. The more ordinary and specific, the harder it is to guess.

The rule is simple: if a voice on the phone says "it's mom / dad / your son / your daughter" and does not use the word, you hang up and call that person back directly on their own number. That is it. AI can clone the voice. It cannot know your family dog is called Doubao.

Five minutes at the dinner table. It could save you a lot of money. More importantly, it could save you a truly awful ten minutes of imagined loss.

Gates said something else in that essay that belongs in your family literacy.

After the Nigerian prince example, he wrote: "people are capable of learning not to take everything at face value." That "capable" is not automatic. It has to be taught. You are the teacher.

At the dinner table, in the car, when you and your kid are watching short-form video together, you can practice three questions with them. These are the family fact-check basics for the deepfake era.

One: where did this thing come from? Is it a first source, or a screenshot of a screenshot of a screenshot? Click on the account that posted it. When did they start posting? Do they only post this kind of content?

Two: who benefits if we believe this? A social account chasing traffic? A company selling something? A political camp who wants us to believe a certain story?

Three: if we wanted to prove it was fake, what would we look up? Is there a reverse image search? Is there original footage? Did this person really say this line? Which interview, which speech, which date?

Do not turn this into a lecture. Do it phone in hand.

You scroll onto something suspicious. You say, "look at this, let us check if it is real." And you check, in front of them. You will teach them many things. But the most useful thing you teach is what a responsible adult looks like when handling information.

That memory of "watching mom and dad do this" is worth a hundred lectures about "do not believe everything on Weibo." It will follow them their whole life.

Finally.

You might think deepfakes are for "experts and governments" to deal with. The reverse is true. Scammers do not attack the White House first; they attack unprepared families. A family code word plus three questions is a guardrail you can install tonight. Simple, unglamorous, effective.

### Questions to sit with

1. Does your family already have a code word only the family knows? If not, at dinner tonight, who do you think will suggest the best word?

2. Of your child's five most-used apps, how many involve back-and-forth with a real person, and how many are one-way content feeds? Is the ratio the one you actually want?

3. The last time a suspicious piece of content came across your feed while your child was near, did you check it in front of them? If not, how do you want to handle it next time?