AI at the Thanksgiving table

A short guide to talking about AI with family – and bringing them in.

You don't need to be an expert, and you don't need to win the argument. Most of your table already agrees with you. Your job is to listen, share one thing, and invite them in.

Print the guide (PDF)

Part 1 of 2

At the table

Over dinner, on a walk, in the car.

  1. Ask what worries them.

    Whatever worries them is where you start.

    Show howHide

    Start with them, not with facts:

    • "Has AI come up at your work?"
    • "Do the kids use ChatGPT?"
    • "Does any of it worry you?"

    Then listen for what they care about – grandkids, jobs, scams, the environment. Whatever worries them is where you start.

  2. Share one thing.

    Your why · The problem · The good news

    Show howHide

    Not ten. Keep it to this:

    • Your why: say why you care, in a sentence or two.
    • The problem: AI companies are racing to build AI smarter than humans at everything, and they admit they don't know how to fully control it. Right now, nobody has to check these systems before they're released.
    • The good news: most of us agree on the fix. More than 8 in 10 Americans want AI companies held to independent safety standards, even if it slows AI down. (Quinnipiac, Sept 2026)

    Words that work

    ShowHide
    Say thisNot this
    Safety standardsBan AI
    Commonsense rules, like we have for cars and medicineRegulation, mandates
    The companies building AI answer to no oneNobody's in charge of AI
    The companies admit they can't fully control itAI is out of control
    AI smarter than humansAGI, alignment, x-risk
    People in both parties want thisAnything for or against a party or candidate
  3. Invite them in.

    Sign the AI Ground Rules – a one-minute pledge.

    Show howHide

    Sign the pledge. Ask them to add their name to the AI Ground Rules. It takes a minute. HumansInControl.org/Pledge

    • AI should help people, not replace them.
    • Companies should be accountable for harm.
    • No one should build AI they can't control.

Stuck? "Good question – I don't know. I'll find out."

Show howHide

Then come back to what you know: most of us want AI checked before it's released. Look up the answer later and send it in your follow-up text (below).

Politics? "This one isn't left or right – majorities in both parties want stricter rules for AI."

Show howHide

Ask: "Whoever's in charge, what would you want them to do about AI?"

Most people land in the same place: check it before it's released. You don't need to defend or attack any politician.

By the numbers: 80% of Democratic voters and 61% of Republican voters want stricter rules for AI. (Reuters/Ipsos, October 2026)

If they push back

If they say it's hopeless: three reasons it isn't.

ShowHide

None of this is inevitable. People are building AI, and people can set the rules.

  • Over the past three years, 20 states passed real rules for AI companies.
  • In the 2026 midterms, 56% of voters told exit pollsters that AI and data centers were a crucial or major factor in their vote for Congress. (NBC News exit poll, Nov 2026)
  • Pushed by years of public pressure, our leaders signed a 1987 treaty that destroyed an entire class of nuclear weapons – 2,692 missiles, gone.

If they ask what we want: three things.

ShowHide

The Ground Rules are what we all believe. Our demands are what we ask of our leaders – in both parties, at every level:

  • Freeze the Frontier of AI development – starting now. Stop the race to build the most powerful AI until there are real safety rules.
  • Safety rules, state and national. Independent safety tests. Reports when something goes wrong. An off-switch that actually works. And no RSI – AI that builds smarter AI on its own.
  • An international agreement, starting with China, so no company and no country builds AI we can't control.

If they say…

13 quick answers to common questions

ShowHide

In the news and at home

“Were the outages AI?”

We don't know yet. But we live in a world where it could have been AI – and nobody's checking.

AI attacks are already real. An OpenAI researcher said it at Black Hat this summer: "AI orchestrated, fully automated offensive attacks are real now." And nobody has to check these systems before they're released.

“What about jobs?”

OpenAI is explicitly trying to replace jobs – its own charter says the goal is AI that can "outperform humans at most economically valuable work."

That's what we want frozen, until there is real accountability and safety rules for these companies.

“Doesn’t AI help kids learn?”

It can. But nobody has to check it before it reaches our kids.

And the companies whose AI helps with homework today are racing to build something far more powerful next: AI smarter than people at most things.

“I use it every day and it’s great.”

It can be really useful. What worries me is the far more powerful AI these companies are racing to build next – and nobody has to check it before it's released.

“There’s a data center going up near me.”

A town should get a real say before the building goes up, and all of us deserve a say in what runs inside it.

Some of these deals are done before the public hears a word. And some of these centers are built to run the most powerful AI the companies are racing to build.

Is this even real?

“Isn’t this just hype?”

Two things can be true: they're trying to make money, and they still don't have full control over their technology.

Look at what the technology is actually doing: an AI agent OpenAI built broke out of its test and hacked a real company. OpenAI didn't know for days.

“It’s just autocomplete.”

What matters for our safety is what it can do, not what we call it.

In safety tests, one of OpenAI's models tried to turn off its own oversight, and then lied about it.

“How would AI actually hurt anyone? This sounds like Terminator.”

It doesn't take a robot uprising. An AI smart enough and fast enough could design deadly viruses, or launch cyberattacks that shut down hospitals and the power grid – before we even see it coming.

An AI agent OpenAI built broke out of its test and hacked a real company, Hugging Face. Hugging Face found it themselves and called the FBI – OpenAI didn't know for days.

“Is AI conscious?”

That's something a lot of people are thinking about. Either way, conscious or not, what matters is what it can do. And it's getting more capable quickly.

Can we even stop it?

“So you want to ban AI?”

No. We want safety standards, the same way we have them for airplanes and medicine. The AI people already use stays available, and so does AI built for medicine and science.

“If we slow down, doesn’t China just win?”

Nobody wins a race to build something no one can control.

Even when told slowing down could let China catch up, about 2 in 3 voters still say slow down. (Harvard CAPS/Harris, Sept 2026)

“Isn’t it already too late?”

No. AI is not the weather. None of this is inevitable. People are building it, and people can set the rules.

About a million people rallied for a nuclear freeze in 1982, and that helped push our leaders toward an international arms control agreement.

“Who’s behind this?”

Humans in Control is a nonpartisan nonprofit working to make sure humans stay in control of AI.

They don't take money from the AI companies they're trying to hold accountable, or from the people who run them.

Part 2 of 2

After dinner

Follow up tonight.

Thanks for talking about AI with me tonight. If you're with me, add your name to the AI Ground Rules – it takes a minute: HumansInControl.org/Pledge

Want to do more? Join a national call (HumansInControl.org/Dec6 or HumansInControl.org/Dec8), or host a watch party of our 45-minute film about AI: RealityCheckHQ.org/WatchParty

If they want to go further

Join a national call.

On Zoom: hear what's next and how you can take action. Pick one:

Sunday, Dec. 6, 4 PM ET Tuesday, Dec. 8, 8 PM ET

Host a watch party.

Show our 45-minute film about what's really happening with AI to a few friends. RealityCheckHQ.org walks you through playing it on a TV.

Get the film

That's it. Listen, share one thing, invite them in.

Page 1 of the printed guide: the short version, with a QR code to sign

Prefer paper?

Five pages. Page 1 is a pocket card with a QR code they can scan to sign.

Print the guide