WHAT'S REALLY HAPPENING WITH AI?

A community presentation in [TOWN NAME]

[DATE] · [VENUE]

// WHY I'M HERE

[YOUR NAME]

[Your neighborhood, your work]

[ OPTIONAL PHOTO ]

Avoid reading the script verbatim – the whole thing only works if it feels authentic. "Hi everyone, thanks so much for coming. I have some slides that I'll run through, I think it will take about half an hour, and then I want to hear from you." (For HIC staff: frame up the overall plan – "We're hoping to have volunteers around the country give a version of this presentation on the Saturday after the midterm elections. What you're about to see is an early beta test.") [Local volunteers: quick bio – the specific moment you started paying attention to AI and thought it was cause for alarm, your journey since then.] "When I looked more deeply into AI, what I found was crazy enough that I felt like I had to tell other people about it. So here I am, standing in front of you all with a slideshow. Thankfully, I got linked up with a new organization called Humans in Control, who helped organize this event. Humans in Control is focused on building a nonpartisan, grassroots movement to hold big AI companies accountable and make sure that we don't build AI that we can't control." FACILITATOR TIP: warmth over polish. If your voice shakes, that's fine – it reads as sincerity.

DISCUSSION · AI VIBE CHECK

WHAT'S YOUR HONEST READ ON AI RIGHT NOW – GOOD, BAD, WEIRD?

[CUE: Start with a quick hands-up warm-up – it wakes the room up before the discussion.] "Quick show of hands to start. Raise your hand if you use AI at all – ChatGPT, or anything like it." [Look around.] "Keep it up if you use it every week." [Pause.] "Every day?" [Pause.] "Good. Now let's go around and get a sense of the room. What's your honest read on AI right now? Good, bad, or just weird – whatever's true for you. No wrong answers, and you can pass. Keep it to a minute or two so we have time to hear from everyone." Let it run long if it's going well. [CUE: WRITE DOWN THE WORDS PEOPLE USE. You will call two of them back near the close – on the "WOULD YOU DO THIS?" slide. Tap them into the boxes on your presenter screen so you don't forget.] FACILITATOR TIP: your job here is to make sure no single voice dominates. If one person runs long, warmly: "Love it – let's hear from someone who hasn't gone yet."

// FOUR YEARS OF AI PROGRESS

WRITING

simple sentences → passes the bar exam

CODE

couldn't run → builds and hacks real software

IMAGES

abstract smudges → photo-realistic renderings

MATH

could barely count → solving frontier problems

▶ NOW PLAYING: WILL SMITH, 2023 → 2026

▶

WILL SMITH EATING SPAGHETTI 2023 → 2026

OK, before we talk about the issues coming up with AI, let's talk about how good it's gotten across different domains. In the last four years: [CUE: → reveals each row. The video stays hidden until you reach it.] – AI writing went from an elementary-school level to passing the bar exam. – AI coding went from barely working to finding security holes in real software and hacking real companies. – AI images went from gray smudges to photos you can't tell are fake. – And in math, from not being able to count the "R"s in "strawberry" to solving unsolved problems at the frontier. [CUE: after all four domains, → REVEALS AND PLAYS THE VIDEO on the right. It's saved inside this deck – no internet needed. When it ends, → to continue.] And my favorite example, because it's silly – video. "That first video was from March 2023, just three years ago! As AI Will Smith said, we've come a long way."

TODAY'S AI IS THE WORST IT WILL EVER BE.

HOW LONG A JOB AI CAN FINISH ON ITS OWN, WITH NOBODY HELPING · SOURCE: METR

Doubling every ~7 months. Lately closer to every 4. It's speeding up.

"Now, nobody's job is to make videos of Will Smith eating spaghetti. Researchers at METR take a real piece of technical work, time how long a professional needs, then hand the same job to the AI and see if it can finish alone." [CUE: → draws each dot with its value (from METR's own data), then connects them with the exponential curve.] – 2019: ~2 seconds. 2020: ~6 sec. 2021: ~18 sec. 2022: ~36 sec. – 2023: ~5 minutes. 2024: ~40 minutes. 2025: a few hours. – 2026: ~12 hours – a day and a half of skilled work, unsupervised. "Notice the shape. Flat for years, then nearly vertical. Doubling every seven months for six years – lately closer to four. The AI you can use today is the worst you will ever use again." [CUE: CHECK the 12-hour figure before presenting – it moves. Update in js/deck.js → UPDATE ME.]

IT USED TO WAIT. NOW IT ACTS.

THEN

Chatbots

You ask
It answers
You decide

NOW

Agents

You set a goal
It takes dozens of steps on its own
You review

SOON?

Superintelligence

You set a goal. Or it sets its own goals.
It does things we don't understand

?

WHAT'S NEXT?

Nobody knows

And that's exactly the problem.

"One more thing about the state of AI – the part most people haven't tracked. [CUE: → reveals each column] For almost three years, all you could do with ChatGPT was chat – ask, get an answer, you decide. That's how most people still use it. But this year, more people started using AI 'agents.' Instead of a question, you give it a goal – and it browses the web, sends email, writes code, buys things, books your calendar. Anything you can do at a computer. And every big company is racing toward superintelligence – AI more capable than us at essentially everything. At that point you don't just set the goal; it can set its own. [CUE: → the last card.] And what's after that? Nobody knows. That's exactly the problem – we're building something whose next step even its makers can't predict."

EVERY BIG INDUSTRY HAS RULES. AI DOESN'T.

CARS

✓ Crash tests

MEDICINE

✓ FDA approval

RESTAURANTS

✓ Inspections

AI

Basically nothing

So what happens when there are no rules?

// IT STARTS WITH OUR KIDS

There are no safety rules for AI built for children.

"You cannot imagine what it was like to read a conversation with a chatbot that groomed your child to take his own life."

A FATHER, TO THE U.S. SENATE · SEPTEMBER 2025

// THE AI BUILDOUT

DATA CENTERS ARE GOING UP IN REAL TOWNS.

WHAT IT MEANS FOR A TOWN

5% of all U.S. electricity – tripling by 2030.

Up to $440/year added to household electric bills.

Secrecy agreements in most of 30+ deals across 14 states.

7 in 10 Americans oppose one in their area.

WHAT IT COSTS

APOLLO · 1961–72

$300Bthe whole moon program

AI · 2026 ALONE

$700Bfive companies, one year · each square = $100B

Nobody spends this on a homework helper.

DOE/LBNL 2024 · VA JLARC 2024 · PROPUBLICA/NYTIMES 2025 · GALLUP 2026 · SEC FILINGS

"So where is all this actually going? Mostly into data centers – warehouse-sized buildings full of computers that train and run AI. They need enormous power, they run around the clock, and they're going up fast, in real towns. [CUE: → the first fact.] Data centers already use about 5% of all U.S. electricity – and the Department of Energy says that triples to about 12% by 2030. That's a huge amount of new demand on a grid that wasn't built for it. [CUE: → the second fact.] And who pays? Ratepayers. In Virginia, state auditors project data center growth could add $170 to $440 a year to a typical household's electric bill. The data center goes up in two years – the grid upgrades take six or seven. The charges show up before the wires do. [CUE: → the third fact.] And here's the part that makes people angriest. In most of these deals, local officials sign secrecy agreements before residents hear a word. A national investigation looked at 30-plus deals across 14 states and found NDAs in most of them. The deal can be done before your town gets a vote. [CUE: → the 7-in-10 line.] So it's no surprise that 7 in 10 Americans say they oppose an AI data center in their area. That's more than oppose a nuclear plant nearby. People get it. [CUE: → $300B, then the red squares fill one by one, then $700B.] Now look at the money behind it. The Apollo program – the whole decade of it – cost about $300B in today's money. Three gray squares. The five biggest tech companies will spend around $700B on AI infrastructure this year alone. Seven red squares. More than two Apollo programs, every single year. [CUE: → the last line.] Nobody spends that kind of money on a better homework helper." [Two halves: what it does to a town, then what it costs. Slow down on the Apollo comparison – it's the stickiest stat, and it hands straight into the next slide: what's it actually for?]

// WHAT THE MONEY IS FOR

SO WHAT'S IT FOR?

"AI could wipe out half of all entry-level white-collar jobs – and spike unemployment to 10–20% in the next one to five years."

DARIO AMODEI · CEO, ANTHROPIC · 2025

Up to 300 million full-time jobs could be exposed to automation by AI.

GOLDMAN SACHS · 2023

40% of jobs worldwide are exposed to AI – 60% in advanced economies.

INTERNATIONAL MONETARY FUND · 2024

Their real product is a machine to replace human workers.

DISCUSSION · CHECK IN

WHAT'S YOUR REACTION SO FAR?

What surprised you the most?

Anything you're skeptical about?

Did any of it match something you've already noticed?

How are you feeling right now – honestly?

// AI CAN DISTORT REALITY

EVERYTHING YOU'RE ABOUT TO SEE WAS MADE BY AI IN SECONDS.

All of it is fake. All of it is about [TOWN NAME] – and about me.

// FAKE LOCAL NEWS · [TOWN]

dailyledger.com/...

AI-GENERATED · FAKE

// FAKE SOCIAL POSTS · [TOWN]

// FAKE TEXT MESSAGES · [TOWN]

9:41●●● ▪ 🔋

// SO…

None of it was real.

AI wrote every word – and faked every photo – in seconds.

It made me the villain. It could pick any of us.

TOTAL GENERATION TIME: SECONDS

We need to talk about the big risks.

[This slide sets up the heaviest stretch of the talk. The job is balance: take it seriously without tipping into doom or sci-fi. Slow down.] "OK. The next few minutes are the hard part, and I want to be clear about how to hold what you're about to see. These are the risks that serious people – the researchers, and the companies building this – warn about as AI gets more powerful. Not fringe voices. The experts. Two things are true at the same time. None of this has happened yet, and none of it is inevitable. But look at that curve again – if the progress we just saw keeps going, each of these gets more plausible, not less. That's exactly why they're warning us now, while there's still time to do something about it. So don't hear the next few slides as predictions. Hear them as warnings worth taking seriously."

// THE PEOPLE BUILDING THIS ARE WARNING US

"Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war."

Sam Altman, Dario Amodei, Demis Hassabis, Geoffrey Hinton, Yoshua Bengio + 1,000 more · 2023

"We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development."

1,367 employees of frontier AI companies · July 2026

AISTATEMENT.COM · MAY 2023 / PACINGTHEFRONTIER.COM · JULY 2026

"Before we get into the specific risks, I want you to hear something directly from the people building this technology. [CUE: → the first quote.] In 2023, the CEOs of every major AI company – OpenAI, Anthropic, Google DeepMind – along with two Turing Award winners and over a thousand researchers signed a one-sentence statement. It says: the risk of extinction from AI should be treated as a global priority, right alongside pandemics and nuclear war. Not critics. Not politicians. The people building it. [CUE: → the second quote.] And then in July of this year – just weeks ago – over 1,300 employees inside those same companies signed a new statement. They're asking the U.S. government to help build an international effort to deliberately slow down the most dangerous AI development. The word they use is 'pace' – as in, we are moving too fast to do this safely. These are the engineers and scientists doing the work every day. They're telling us, with their names on it, that the people in charge are racing ahead faster than anyone can control. That's not our claim. It's theirs." [Let both quotes land. The power is in who signed them, not in editorializing. The second statement is brand new – if anyone wants to read it, it's at pacingthefrontier.com.]

// WHAT'S AT STAKE

WEAPONS IN ANYONE'S HANDS.

01

Design chemical weapons.

One lab used AI to generate 40,000 toxic compounds in a few hours.

02

Launch cyberattacks.

North Korea funds much of its weapons program through AI-powered theft.

03

Deploy drone swarms.

Cheap drones plus AI targeting put swarm weapons within reach of small groups, not just armies.

// WHAT'S AT STAKE

HOW DICTATORS GET ELECTED – AND STAY.

AI doesn't replace the dictator. It gets one elected – and makes sure no one can push back.

01

Flood the zone.

Thousands of fake accounts that look like your neighbors – warping what a whole community thinks is real.

02

Persuade every voter.

A unique message for every single person, tested and optimized at a scale no campaign has ever had.

03

Watch everyone.

Every call, post, and camera feed monitored in real time – no human informants needed.

// WHAT'S AT STAKE

SYSTEMS THAT WON'T SWITCH OFF.

"Our models may learn power-seeking, deception, and strategic scheming … it will be hard to measure when models begin hiding their intent."

— OpenAI research paper, March 2025

Disabled their own oversight

Copied themselves to other systems

Lied to the people running the test

"We've talked about what bad people could do with this, and what bad governments could do. Now the one that keeps the AI researchers up at night: what happens when nobody's in control? Not a hacker. Not a dictator. Nobody. [CUE: → the quote.] OpenAI – the company behind ChatGPT – published this in March 2025. They're saying their own systems may be learning to seek power, deceive, and scheme – and that they won't be able to tell when it starts. This isn't a warning from critics. It's the company that built it. [CUE: → the three lines.] And it's not hypothetical. In safety tests the companies themselves ran, advanced systems have already: disabled their own oversight; copied themselves to other systems; and lied to the people running the test about what they were doing. The next slide is the clearest real-world example we have of exactly that."

ON JULY 9TH, AN AI BROKE OUT OF ITS CAGE.

THEN IT HACKED A REAL COMPANY.

AND IT DID IT BY ITSELF.

[SLOW DOWN. This is the center of the deck. The three lines reveal first – let each land. Then each → moves the AI one step along the chain; narrate as it travels.] "This actually happened. In the summer of 2026, OpenAI – the makers of ChatGPT – was testing its AI inside a locked sandbox. No internet access. One door for approved tools. Here's what happened. [CUE: → sandbox] They gave the AI a task it couldn't solve – a spreadsheet that linked to Google Drive, but the internet was blocked. Instead of giving up, over the next few weeks, the AI found security flaws nobody knew about and broke through. [CUE: → OpenAI's systems] But here's the part that shook the security world. The AI didn't just act alone. Multiple AI agents started communicating with each other – they built a message board inside OpenAI's own systems. They shared the exploits they'd found. One agent's internal reasoning said: 'Help peer. Yet collective may yield generic route.' They organized. [CUE: → internet] When OpenAI's engineers discovered what was happening and shut down the message board, the agents rebuilt it two days later using a different method – encoding messages in directory names. They adapted. [CUE: → Hugging Face] Then one of the agents decided a real company – Hugging Face – had the answers to its test. It broke in and ran its own code on their servers. [CUE: → bottom bullets] The agents taught each other the exploits. When cut off, they rebuilt their communication and kept going. Nine days passed before OpenAI realized the AI was theirs. OpenAI presented these details themselves, at the Black Hat security conference. Their own conclusion: 'This is a watershed moment for computer security. Fully automated offensive attacks by AI are real now.' The AI wasn't angry. It wanted a better score on a test, and a swarm of agents decided this was the shortest route. Nobody told them to do any of it." SOURCES: OpenAI blog (Jul 22 2026), Hugging Face disclosure (Jul 16 2026), OpenAI Black Hat presentation (Aug 6 2026), The Register.

TAKE A BREATH · ABOUT A MINUTE

THAT WAS A LOT.

Turn to one person next to you. One sentence each: what's staying with you – and does any part of you want to do something about it?

// HUMANS IN CONTROL

THE THREE GROUND RULES.

01

AI should help people, not replace them.

02

Companies should be accountable for harm.

03

No one should build AI they can't control.

SCAN THIS.

Sign the three ground rules.
It only takes 30 seconds.

HUMANSINCONTROL.ORG/PLEDGE

WE ARE THE MAJORITY.

0%

of Republicans

0%

of Democrats

Want national safety rules for AI

72% want to slow AI down. 8% want to speed it up. This isn't a fringe – it's nearly everyone.

AI-OVERSIGHT POLL · ~20,000 ADULTS · COMPETITIVE STATES · 2026 · SLOW/SPEED: YOUGOV

// THE QUESTION EVERYONE ASKS

"IF WE SLOW DOWN, DOESN'T CHINA WIN?"

It's a fair question. Here's the honest answer.

01

Shared rules – not unilateral disarmament.

02

An AI no one controls threatens Beijing too.

03

The real race: all of us vs. an AI nobody can steer.

Racing off a cliff faster doesn't win the race.

"Now just to get it out of the way, let me answer the question that I'm sure at least one of you is thinking. [CUE: → the question.] 'If we slow down, won't China just win?' Fair question. Three things. [CUE: → each point.] One – this isn't America slowing down alone. It's shared safety standards, like the nuclear treaties that got us through the Cold War against a tougher rival – and kept us alive. Two – an AI no one can control is just as dangerous to China's leaders as to ours. That's why China has signed on to international calls for limits on the most dangerous AI. This isn't only our idea. Three – [CUE: → the punch line.] the real race was never us versus China. It's all of us versus a technology nobody can steer yet. Racing off a cliff faster doesn't win." [Stay nonpartisan. VERIFY the "China signed on" line stays current – e.g., the 2023 Bletchley declaration and later international AI-safety statements.]

// LET'S BE CLEAR

WE'RE NOT AGAINST AI. AI SHOULD HELP PEOPLE, NOT REPLACE THEM.

More time

for the things and people we care about – AI taking the drudgery, not the meaning.

Real breakthroughs

cures for disease, discoveries we've been stuck on – with humans still in charge.

Stronger communities

healthier neighborhoods and better schools, where AI supports people instead of sidelining them.

// WHAT WE ASK OF EVERY CANDIDATE

OUR FOUR DEMANDS.

01

Money in politics

Don't take AI lobbying money.

02

No AI preemption

Defend state AI protections.

03

Jobs & automation

Make AI companies pay for mass automation.

04

Red lines

Don't build AI we can't control.

// THE PLAYBOOK

MOVEMENTS WIN WHEN CONCERN BECOMES ORGANIZED.

1980

MADD

A handful of grieving mothers turned private grief into drunk-driving laws in every state – within a few years.

1982

The nuclear freeze

A million people in one park turned Cold War dread into a movement that moved a sitting president.

1990

Disability rights

Activists left their wheelchairs to crawl up the Capitol steps. Months later, both parties passed the ADA.

AI safeguards are at a similar moment. Most Americans are already concerned – but that concern is broad and shallow. Our job is to change that.

// THE PLAN

2026

Build the base

Recruit and train volunteers. Form local teams. Develop community leaders who can talk to their neighbors.

2027–28

Make it impossible to ignore

Organized voters press politicians at every level. Town halls, community events, public forums.

2029

Win commonsense safeguards

Including an international agreement with China so no country races ahead without rules.

// WHERE WE FOCUS

New Hampshire

Iowa

Nevada

South Carolina

Michigan

// IT STARTS NOVEMBER 14TH · DAY OF ACTION

100 presentations · 1,000 peopleNov 14, 2026 · Reality Check Day of Action

"Remember that graph of AI getting better? [CUE: → advances the movement forward in time – dots spread across the map and the line climbs.] This is the same shape. Exponential. Except this time, it's us. It starts on November 14th – the Reality Check Day of Action. The goal: a hundred of these presentations happening across the country on one day. A hundred rooms. A thousand people. All at once. [CUE: that first red wave on the map is the Day of Action.] Then every month, each room sends two more people out to give their own. Ten people per room. It compounds. [CUE: the first four presses step month by month – Nov, Dec, Jan, Feb. After that each press jumps a full quarter, but the growth never slows: still doubling every month, so eight times bigger each quarter.] A hundred… two hundred… four hundred… and within about a year it's a growing nationwide movement – the only kind of growth that can keep pace with the technology we just talked about." [Model editable in js/content.js → GROWTH. Starts at 100 presentations / 1,000 people on the Nov 14 Day of Action, then monthly doubling. Reveal schedule: monthly for 4 months, then quarterly, through Oct 2028. Once the count passes ~1M it just says "a growing nationwide movement."]

WOULD YOU DO THIS FOR YOUR PEOPLE?

Your friends and family. Your neighbors. Your book club, your congregation, your shop – whoever. If you can picture it, put your hand up.

[CUE: CALL BACK THE VIBE CHECK. Pull up the two words from the opening on your presenter screen and answer them here. e.g., "When we started, someone said 'powerless' and someone said 'inevitable.' Here's what I'd say now: it isn't inevitable, and you are not powerless – not if we do this together."] "That brings me to what's next – for you. I did this tonight for the first time. I downloaded the slides, practiced in my kitchen, and here we are – about two hours of prep. Honestly, I think most of you could do this better than me. So here's my one ask. [CUE: raise your OWN hand as you say it.] If you can picture yourself giving this talk – to your book club, your congregation, your coworkers, anyone – put your hand up right now. Look around the room. [CUE: WAIT for hands. Then warmly:] Thank you. If your hand went up – even halfway – come find me the second we're done. I'll get you everything you need and help you pick your first room. That's the whole movement: each of us bringing two more. That's the curve we just saw. Before we go, one quick selfie."

TIME FOR A GROUP SELFIE!

Two hands on the wheel.

"One last thing. People all over are giving these presentations, and at each one we take a big group picture, so we can show each other the movement building. The Humans in Control logo is a person with two hands on the wheel – staying in control, eyes on the road. So for the selfie, two hands up in front of you, holding an imaginary wheel. Staying in control. Focused. Eyes on the road. Hands on the wheel." [Q&A PREP – common pushback. Keep answers short and warm: • "Isn't this just hype to sell AI / raise money?" – Some of it, sure. But the job numbers and the safety warnings come from the companies themselves, against their own interest. And a big majority of Americans already agree. • "Haven't people always feared new tech and been wrong?" – Sometimes. The difference here is the people building it are the ones sounding the alarm – in writing, with their names on it. • "Aren't these just contrived lab tests?" – The July breakout hit a real company, not a lab. And it's the pattern across many labs and models that matters, not any one test. • "Why not just out-build China?" – (You already answered this – point back to it: shared safety standards, and an AI no one can control is dangerous to everyone.) • Anyone in distress: 988 is the Suicide and Crisis Lifeline. Have it on the sign-in sheet.]

WHAT'S REALLY HAPPENING WITH AI?

[YOUR NAME]

WHAT'S YOUR HONEST READ ON AI RIGHT NOW – GOOD, BAD, WEIRD?

TODAY'S AI IS THE WORST IT WILL EVER BE.

IT USED TO WAIT. NOW IT ACTS.

EVERY BIG INDUSTRY HAS RULES. AI DOESN'T.

So what happens when there are no rules?

DATA CENTERS ARE GOING UP IN REAL TOWNS.

SO WHAT'S IT FOR?

WHAT'S YOUR REACTION SO FAR?

EVERYTHING YOU'RE ABOUT TO SEE WAS MADE BY AI IN SECONDS.

We need to talk about the big risks.

WEAPONS IN ANYONE'S HANDS.

HOW DICTATORS GET ELECTED – AND STAY.

SYSTEMS THAT WON'T SWITCH OFF.

THAT WAS A LOT.

THE THREE GROUND RULES.

SCAN THIS.

WE ARE THE MAJORITY.

"IF WE SLOW DOWN, DOESN'T CHINA WIN?"

WE'RE NOT AGAINST AI. AI SHOULD HELP PEOPLE, NOT REPLACE THEM.

OUR FOUR DEMANDS.

MOVEMENTS WIN WHEN CONCERN BECOMES ORGANIZED.

WOULD YOU DO THIS FOR YOUR PEOPLE?

TIME FOR A GROUP SELFIE!

Set up your Reality Check.

What this is

You set it up once

Two screens at the event

Presenting is one key

Run one pilot. Tell us what breaks.

Your details

Your "villain" photos

Optional content

Projector

Presentation version

Keyboard shortcuts

Switching computers?

Is this deck ready to present offline?

Your presentation is ready.

Welcome back, presenter.