Return to Transcripts main page

Smerconish

Ex-Anthropic Researcher: AI Could "Kill Us All By The End Of The Decade." U.S. Commemorates 25th Anniversary of 9/11. Aired 9-10a ET

Aired September 12, 2026 - 09:00   ET

THIS IS A RUSH TRANSCRIPT. THIS COPY MAY NOT BE IN ITS FINAL FORM AND MAY BE UPDATED.

(COMMERCIAL BREAK)

[09:00:26]

MICHAEL SMERCONISH, CNN ANCHOR: So how exactly would AI kill us? I'm Michael Smerconish in the Philly burbs. Some lawmakers consider the threat of artificial intelligence so pressing that they recently sent a letter to House Speaker Mike Johnson. They urged him to cancel the House recess so that they can pass protective legislation. You know they're serious when they're talking about cutting short their vacation.

Lawmakers on both sides of the aisle have been speaking out on the issue lately, especially because a major warning from people who would know has now hit the headlines. For several days we've been hearing from two individuals on the inside of Anthropic, one who quit, the other who remains, both warning that the technology they help build could threaten humanity.

Jacob Coxon, a 27-year-old former Anthropic researcher, announced his resignation this week in a series of posts on X. He wrote this, "The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything, many executives and senior researchers will couch their phrasing in the press to sound sensible, but I hear the same people express fear privately. No other human activity poses this level of danger."

Then Evan Hubinger, a senior Anthropic employee who leads research on controlling future AI systems, publicly backed him up. Hubinger wrote this. He said, "Jacob is correct here, we really do earnestly believe AI could kill all humans. I personally think it's greater than 10 percent within the next decade. I believe Anthropic is trying its best, but we do not yet have a plan to solve alignment for superintelligence and are not clearly on track to."

Those are extraordinary warnings, but they leave an obvious question. How would it actually happen? Well, the Wall Street Journal breaks the fear into two broad categories, human misuse and loss of control. First, human misuse. A bad actor could use an advanced AI system to design a novel virus, develop a biological weapon, or launch a massive cyber-attack.

And that risk is no longer entirely hypothetical. In a report released on Thursday, Anthropic said that it blocked multiple accounts that used its models in ways that could support the development of biological weapons. Over 30 days, the company identified about 35 separate research efforts involving potentially concerning activity, including work involving bird flu, other infectious diseases, venoms and toxins. Importantly, Anthropic could not determine whether those users intended harm or were conducting legitimate scientific research.

The second fear is loss of control. Not the terminator, not robots suddenly deciding that they hate us. The concern is that a super intelligent system could pursue an assigned goal without regard for human life. Researchers call that misalignment. The Wall Street Journal article notes further that in lab tests, some AI models have displayed power seeking behavior, attempted to copy themselves, avoided being shut down, and try to conceal what they had done.

Take the classic thought experiment. Tell a super intelligent machine to make as many paper clips as possible. If it pursues that goal without human values or limits, it could eventually treat everything on earth, including us, as raw materials for more paperclips. The machine does not need to hate humanity. It just needs to stop caring whether we survive.

Which brings us to another question. If the people building these systems believe the system is the real risk, then why keep building them? Anthropic and OpenAI say the benefits could be enormous, from developing new medicines to advancing scientific research. They also argue that superintelligence may be inevitable and that the United States cannot allow an authoritarian government to get there first. Anthropic says that its safeguards are working, but acknowledges that stronger protections will be needed as the models become more powerful.

OpenAI's chief scientist says no lab has yet solved the problems of monitoring and controlling such systems.

Now there are skeptics. Some say these warnings help AI companies market their products as incredibly powerful. Others say calls for regulation could help the biggest companies lock out the smaller competitors. But the warnings keep coming from the people building the technology. The White House has asked leading developers to voluntarily submit certain models for testing before release.

[09:05:00]

And members of both parties have proposed kill switches and mandatory reporting of serious safety incidents. So far, those proposals have gained little traction. So we're left with a race toward a technology that could transform human life led by companies that say they do not yet know how to control its most advanced form. Is that responsible progress, or are we building first and hoping to find the brakes later?

It all leads me to today's poll question at smerconish.com, could I wipe out humanity? Joining me now is the nation's first cyber czar, Richard Clarke. He served for 30 years in national security roles in the U.S. government, including 10 years in the White House under three different American presidents. Currently, he's the CEO of Good Harbor Security Risk Management.

OK, Richard, March 24, 2004 your words to the 9/11 Commission we all watched on national T.V., you said your government failed you. Those entrusted with protecting you failed you. You said I failed you. Do you see parallels between pre September 11th and this conversation about AI?

RICHARD CLARKE, FORMER WHITE HOUSE COUNTER TERRORISM ADVISER: Michael, there are enormous parallels. In both cases, we had experts in the field giving clear warnings of an impending disaster, warnings based on empirical evidence. And they were ignored or little was done to address their concerns.

I think saying now that humanity is at risk is a little bit hyperbolic. It's meant to scare people. And when he says there's a less than or greater than 10 percent chance of that happening, perhaps. There's no way of knowing. But what we do know is there's probably 100 percent chance of major disasters short of eliminating humanity.

Major disasters happening with AI if we don't do something soon and the window for action is very, very small. We have to do something in the next several months to address these issues that you just raised, the lack of transparency. AI people don't know why their models do what they do. Deception. We know from the AI modelers themselves that these models are being deceitful.

They're hiding stuff from their creators. We don't know what they will do. But we do know they have dangerous knowledge. They know how to make a bioweapon. They know how to make a nuclear weapon.

They shouldn't have that knowledge. That knowledge needs to be deleted. There are technological solutions to these problems for transparency, for deletion of knowledge. And there are regulatory approaches, as you mentioned, monitoring, reporting, kill switches. All of this can be done.

But right now the government is doing nothing except voluntary review for hacking. We have the president of China coming here to Washington on the 24th. He has publicly talked about his concern with the possible loss of control. Wouldn't it be nice if our president talked publicly about loss of control being a concern? Wouldn't it be even nicer if the two presidents could agree to work together to stop loss of control?

SMERCONISH: You know that some will say but we can't trust President Xi. And no matter what he says, if we put the brakes on it'll be advantage China. And to them you would say what?

CLARKE: That I worked for 25 years on arms control where I was told you can't trust people to not build biological weapons, you can't trust people not to build chemical weapons, you can't trust people to regulate the number of nuclear weapons, and we developed over time through negotiations, verification methods that gave us enough certainty that we signed treaties on all those issues and the Senate ratified treaties on all of those issues. It's going to be tough, but you have to start somewhere and you have to start soon.

SMERCONISH: On this same issue, I want you to watch a brief clip of what Senator Cruz said this week. Play it.

(BEGIN VIDEO CLIP)

SEN. TED CRUZ (R-TX): Listen, I -- I've actually joked that if, and this is somewhat tongue in cheek, but it's not entirely if they're going to be killer robots, I'd rather they be American killer robots and not Chinese killer robots.

(END VIDEO CLIP)

SMERCONISH: Richard Clarke, you think what, when you hear and watch that?

CLARKE: I think this is an excuse for not acting. I think this whole China boogeyman, China shibboleth is something we have to get over. The Chinese are concerned about loss of control. They don't want that to occur any more than we do.

And by the way, most of the Chinese models are stealing from American models through a process called distillation. They're not going to outpace us. If we slow down, they will slow down perforce because that's how they develop their models.

[09:10:13]

SMERCONISH: I find it interesting that where the Feds have not acted, where the Congress has not acted, states have taken action. Some say they worry about a patchwork that will result when you've got these 50 different lab experiments. Will you say something about regulation best on a state level or federal government level?

CLARKE: Obviously, it's best on the federal government level. But California and other states, New York can prompt federal action by filling in the gap, by coming up with good, smart, modern regulation, and that will force federal action. But we do need the federal government, we need the Congress, we need smart new regulatory regimes and we need international action. Not just China and the United States, but Europe as well. And until we get that, we are at risk, and that risk is significant.

It may not be the end of humanity, but it's risk of nuclear weapons, it's risk of biological weapons, it's risk of the cyber-attacks wiping out infrastructure. We're going to have disasters, Michael, if we don't do something. That is 100 percent.

SMERCONISH: You said -- you said that the anthropic language, the two employees, one former, one current, was hyperbolic. Short answer, if you can, what worries you most?

CLARKE: What worries me most is the two things you mentioned from the Wall Street Journal, human misuse and loss of control where the AI goes off on its own. We've had ample evidence. This is not science fiction anymore. The AI companies themselves are telling us they've lost control. They don't know what these things are doing.

They don't know when they're out on the Internet attacking websites. They don't know when they're being deceptive. They don't know why they're making the decisions that they're making. That's dangerous. We cannot regulate everything else in the world and not regulate AI that could do the significant damage.

We have regulations for guardrails on highways, but we don't have regulations for guardrails on AI.

SMERCONISH: Before we say so long to Richard, can we put the social media reaction on the screen perhaps? I want to hear his reaction. From the world of X, these AI companies have no idea how to control their models. These models are just babies and are escaping. What happens when they grow up?

Richard Clarke, you would say what to John?

CLARKE: That they do have an idea of how to control them. It's more expensive and they're not going to do it unless everybody in the business has to do it because they'll lose competitive advantage. For example, deleting dangerous knowledge through a process called gradient masking, that costs money. They have to retrain their models and -- in a way that we can delete knowledge about nuclear weapons and biological weapons.

No one company is going to do that. No one company is going to fund that research or implement that research unless everybody has to. And so let's level the playing field required for everybody, not just the hyperscalers, but all open source, open weight models. We have to do it and we have to do it soon.

SMERCONISH: Richard Clarke, thank you as always.

And to everyone at home, I want to know what you think. Go to my website at smerconish.com answer today's rather morbid poll question, could AI wipe out humanity?

What a perfect guest to have after yesterday's 25th anniversary of the events of September 11th. He who played a significant role comes on March of 2004 in front of the 9/11 Commission and says, look, we failed you. And here he is today saying the parallels are real.

Up ahead, President Trump has ordered new restrictions on mail in ballots. He says it'll save the system. So why do so many see it as a form of potential sabotage?

Please sign up for my newsletter at smerconish.com, the daily newsletter. When you're there voting on the poll question -- I love this, from Eric Allie. What a great sketch. So timely to our conversation.

(COMMERCIAL BREAK)

[09:18:33] SMERCONISH: There is no list. By now, you've heard that President Trump signed an executive order with the stated purpose of ensuring that voting in federal elections is limited to U.S. citizens and of using the U.S. mail to enforce that restriction, which sounds reasonable. Only citizens have the right to vote. But how would it work?

Under a rule that the post office finalized late last month, a state that wants to send ballots through the mail must first upload the name, the address, and a unique barcode for every voter it intends to reach into a new federal portal. Ballots that are not in the system get bounced back to the election office, which means the postal service can refuse to carry a ballot to any voter who isn't on the list. Hence, the postal service is now a gatekeeper on who can vote by mail. That might make sense if only the postal service had access to a list of citizens, but it doesn't.

In fact, in a filing at the Supreme Court this week, the government conceded that the postal service has, quote, "no role in verifying whether male voters are eligible," and no one else has such a list either. This has ramifications beyond President Trump's executive order, of course. For example, the president just announced at his convention on Wednesday that if Republicans win the House and Senate, he will issue a dividend to every adult citizen in the United States for $5,000.

Whether or not this is a good idea, how can he be sure that he's going to get all those checks only to citizens and to nobody else? In fact, the United States has never had a comprehensive list of its citizens. That's not an oversight, it's by design.

[09:20:12]

Almost as soon as the Constitution was ratified, the Framers added a Bill of Rights to reassure individual citizens that they had protection against an overreaching federal government. Think about it. The first, the second, the fourth, the Fifth Amendments, arguably even the third, are all about keeping Washington at arm's length. Freedom of speech, of the press, of worship, to petition, to peaceably assemble. Something foundational is at stake when the federal government starts assembling the means to surveil its own people.

And here's the practical problem with quickly building a list. Birth records are maintained by states and counties, not the federal government. There is no single file that reliably identifies every citizen by name, address, and citizenship status. Instead, we have records of subsets of the population. For example, DHS, they have naturalization records for people who became citizens through the immigration process.

The State Department, they have passport records for people who have obtained a passport. And the Social Security Administration has records that include a citizenship field. But those records were designed to administer benefits, not to determine who can vote. So what's the intended solution? The President's Executive Order directs the Department of Homeland Security working with the Social Security Administration to build something the country has never had, a state citizenship list for every state.

A roster of every confirmed U.S. citizen, everybody 18 and over living within its borders, and to deliver it to state officials no fewer than 60 days before every federal election. And then it directs the attorney general to prioritize the prosecution of state and local election officials who send a ballot to anybody who's not eligible. Because the order relies on those partial databases, whole categories of citizens are at risk of being left off or misclassified.

Think native born citizens without a passport or naturalized citizens with outdated or inconsistent records or citizens born abroad to U.S. citizen parents or people whose names changed through marriage, divorce, adoption or a court order, older Americans and anybody who rarely deals with the federal government and Americans with common names, hyphenated names or a junior or the third after their name, anyone a computer might enter two different ways. We don't have to guess how a list built from those sources performs. We already watched it.

The administration expanded a DHS tool called SAVE so that states could run their voter rolls against Social Security data. Texas ran it. They flagged more than 2700 registered voters as potential noncitizens. In Denton County, one in seven of those flagged had already proved that they were citizens by the time that reporters checked. In Boone County, Missouri, more than half of those flagged were citizens.

Alabama inactivated more than 3,000 voters and later acknowledged that more than 2,000 of them were eligible. In June, a federal judge in Washington blocked that overhaul, writing that officials haphazardly combined and repurposed the private information of millions of Americans, including citizenship data that they knew to be unreliable.

And now scale that up. Not a few thousand flag names in one state, but a list of every citizen in every state rebuilt on a 60-day cycle from records nobody designed for this purpose. The administration's own June 30 deadline for the portal that would hold these lists, that came and went. The September 4th target 60 days out from the election, that came and went. NPR reports that the Justice Department now says only that it will give opposing lawyers 48 hours notice before the lists go live.

And meanwhile, close to 100 million mail ballots need to move this fall. The New York Times editorial board they just weighed in on this big picture issue. They claim the order oversteps the president's authority because the Constitution leaves the time, place and manner of elections to the states. The barcode system is rushed. A postal whistleblower warns that one bad scan can reject an entire batch of ballots.

Nearly one in three Americans now votes by mail, and the post office has no experience deciding who gets a ballot. That's all true, and it's typical of the coverage of this issue. Missing is the word citizenship and any acknowledgement that when you get beyond the barcode, there's still the question of how to test eligibility based on whether voters are actual citizens. We'll see if President Trump's plan goes forward, he's certainly getting plenty of pushback. A federal judge in Boston, Indira Talwani, has blocked the administration at every step since June, most recently the postal rule itself. And on Thursday, an appeals court ruled that they would not lift Judge Talwani's preliminary injunction that prevented the post office from carrying out Trump's executive order.

[09:25:21]

Meanwhile, this week, 23 states and the District of Columbia filed a request urging the Supreme Court to let the judge's opinion stand. And at present, were awaiting a decision from SCOTUS.

I raised this issue on my Sirius XM radio program this week with Senator Chris Coons of Delaware. I told him that I worry there will be people whose names won't appear on the list who should, and vice versa, and that I see that as a prescription for chaos. Coons, before he was in the Senate, spent a decade as a county elected official. And he told me that one of the real strengths of our electoral system compared to many other democracies is that it's diffuse, run overwhelmingly by nonpartisan, dedicated local people, volunteers who staff the polls, elected secretaries of state, county election boards, and that keeps any one party or interest from controlling the machinery of the elections. It's a foundational principle that the federal government doesn't run our elections.

And not lost on Senator Coons in any other context, the people pushing for a national database like this, they would be the first ones to oppose it.

(BEGIN VIDEO CLIP)

SEN. CHRIS COONS (D-DE): Look, there has long been a widespread resistance amongst libertarians and conservatives to the federal government having anything like a list of gun owners or a list of citizens or a list of our individual addresses. You know, look, one of the defining features of Communist China is that they surveil, monitor their citizens, their residents in every aspect of their lives. We allow companies to do it. I mean --

SMERCONISH: True.

COONS: -- Google and Meta have huge quantities of data.

SMERCONISH: Yes. They -- I'm sure they know the answer --

COONS: They know the answer.

SMERCONISH: -- but the government doesn't know the answer.

(END VIDEO CLIP)

SMERCONISH: But most importantly, Senator Coons underscored the fallibility of government records.

(BEGIN VIDEO CLIP) COONS: And look, Michael, a great part of my job as a senator on behalf of a million Delawareans is addressing constituent concerns. And I get a summary every single week from my team of what it is people are calling and e-mailing about, they want me to know about or want me to -- and constituent cases. Every single week, somebody reaches out to us about the IRS, Social Security, or Veterans Administration making mistakes. Some of them are astonishing. You know, the Social Security Administration sending a widow a letter saying, we have overpaid you $40,000 over the last decade.

You must repay us immediately. And she calls and calls and can't get anyone on the phone, is trying to figure out what's going on. You know, that's not every day or every week. But we routinely get less dramatic examples, but still regular examples where the VA --

SMERCONISH: True.

COONS: -- IRS or Social Security generates wrong information, drops people from the rolls, demands that people repay something they don't owe, or denies them benefits they've earned. And I have five people in Wilmington, that's all they do every day, all day, and two people in Dover, and that's just a million people. We have 330 million people. How many mistaken records are there at the IRS, at the Veterans Administration, State Department passport list? You know, I think for us to imagine that the federal government with no real preparation for this, it's not like the census where they spend years preparing to carry it out, by the way, census records, it will not go well if there is an insistence by the U.S. Supreme Court on implementing this ill- conceived solution in search of a problem.

(END VIDEO CLIP)

SMERCONISH: Let me be clear about what I'm not saying. I'm not against verifying citizenship. No one who's ineligible should be casting a ballot. But the evidence says that almost nobody's doing that. When seven states ran roughly 35 million registered voters through the federal tool, they turned up about 4200 possible non-citizens, about 1/100th of 1 percent.

And we know from county follow ups that a meaningful share of those flags were wrong. You know there is a fix for this. Michael Lind writing in Tablet, argues that a national ID for citizens would solve illegal employment, non-citizen voting and identity verification in one stroke. More than 170 countries have one. But as a writer at Reason magazine countered, linking every American's records in one federal database also creates, in the words of the late Senator Daniel Akaka, one stop shopping for identity thieves.

Now that's a debate worth having. It's a debate this country has ducked for 250 years and for good reasons that go to the Bill of Rights. What you can't do is skip the debate, order up the list in the 60 days before the records that don't even exist, hand it to the post office and threaten to prosecute the county clerk who gets it wrong.

[09:30:05] The Supreme Court may decide the fate of the postal rule any day, but whatever the justices do about the envelopes and the barcodes, the harder question remains, who is a citizen and how does Washington know? Don't try to check the list because that's impossible. There is no list.

Still to come, more of your social media reaction. And go to my Web site at Smerconish.com, vote on today's poll question. Could AI wipe out humanity?

Plus, the towers fell 25 years ago. We must never forget. But as a nation, could we all come together and unite as we did in the aftermath of 9/11? Or are we all too politically divided today?

Sign up for my newsletter at Smerconish.com. You'll get the work of Jack Ohman among others.

(COMMERCIAL BREAK)

[09:35:34]

SMERCONISH: So, how do you find me in social media? You can follow me on X. You can subscribe to my YouTube channel. All the usual places, Facebook, the newsletter.

Coincidentally, mail-in voting is just like AI. Without guardrails it can be disastrous. We need to verify that the votes cast are legitimate.

Look, it originated in the western states. States like Oregon have no incident. I mean, it's been -- it's been really smooth operating. And yes, there have been a couple handful of screw ups but the data just doesn't suggest that we have a problem.

The data, let me say it clearly, does not suggest that we have a problem that warrants the kind of fundamental overhaul that the administration is seeking to carry out right now. And for all the reasons that I revealed in my commentary, the federal government doesn't have a citizenship list and put the bar codes aside. So much of a -- so much of the attention so far has been on the portal, and the postal service. People love to demonize the postal service.

Let me tell you something. The postal service has an awesome record of delivering ballots. I'm trying to -- I'm imploring you to take your eye off barcodes in the postal service and just assume that they function well. Where's the list? You know, it's going to be a blank show because people who ought to be on it won't be. And people who shouldn't be will be.

The DMV case from Jersey a couple of weeks ago, your most recent example. This is a real issue. Supreme Court is going to deal with it any minute, so hopefully they'll put an end to it.

Again, I don't want to be misunderstood. I want there to be safeguards and ensure that people who are citizens are voting. But this is not the way to do it. OK. Sorry, I was long-winded. Give me another one, please.

I just checked with ChatGPT and it said there's nothing to worry about, says Joe.

You know, Joe, I'll tell you something. Richard Clarke, really the perfect guest to have today, the nation's first cyber czar, he was not playing that role at or about September 11 because we weren't focused on cyber issues but -- like he's the guy who three years after September 11, having served three American presidents in front of the 9/11 commission, comes on in a very emotional moment says, we failed you, I failed you.

And 25 years ago, you know, yesterday was that anniversary. And here he is today on the program saying the parallels are unmistakable. The same flashing lights that we had pre September 11, he sees now relative to the cyber threats that face us.

Please make sure that you're voting on today's poll question at Smerconish.com. I know it's a little gruesome but the "Wall Street Journal," the Anthropic stories this week, it had to be this. Do you think that AI could wipe out humanity?

And still to come, September 11, which was meant to tear us apart, actually unified us at least then. But now that a generation plus has grown up since then, does it mean the same thing? For that matter, could any event ever unify us again?

Please register for my newsletter, it's free and it's worthy, when you're voting on the poll question of the day. I thought this was perfect. I thought this was beautiful. Steve Breen drew this for my newsletter this week. And it just summed up so much. And not lost on me is that he used purple as a background color.

(COMMERCIAL BREAK)

[09:43:09]

SMERCONISH: For 25 years yesterday, I'll never forget where I was, if you're of a certain age, neither will you. Me I was in a courtroom in Philadelphia's city hall, day two of a case that I was trying when a member of the judges staff leaned in and said that a plane had just hit the World Trade Center. I assumed it was an air traffic control mistake.

By that afternoon, I was in a radio studio talking about a man named Bin Laden, whom I could barely have identified that morning. I went back and listened to the tape this week. I don't sound like a person who knew what was coming.

Everybody old enough has a version of that story, where we were, who we called, how long we sat in front of the television. For a quarter century, I've called 9/11 the December 7th of our lives, and I meant it as a measure of scale.

This year, I hear it as a warning because Pearl Harbor was the day that would live in infamy. And it did for a while. By its 25th anniversary in 1966, the country had moved on to Vietnam. The men who were on the Arizona, they never forgot. The country, in a way that countries do, kind of put it in a drawer.

That's not a failure of our character. It's what time does. And what worries me now is how much faster it seems to be happening and why.

Think about the world on September 11, 2001. The internet was something you dialed into over a phone line, and half of us didn't. The iPhone was still six years away. There was no Facebook, no YouTube, no Twitter, no feed.

If you wanted to know what happened, you turned on a television. There were a handful of places that you could turn, and everybody you knew was watching one of them. We pretty much watched the same shows on the same nights and then talk about them the next morning at the office because that's where we went. Nobody was working yet from the kitchen table, and we were not yet fully sorted into our political silos.

[09:45:01]

We were, without ever thinking about it, in each other's company. And that's why the country came together the way that it did. Not because we were better people in 2001, but because we were in the same room.

Remember the flags on the overpasses, the lines at the blood banks, members of Congress singing on the Capitol steps, a 90 percent approval rating for a president who had won with fewer votes than his opponent just 10 months prior? None of that required a plan. It was what a connected people did on reflex. And I'm not sure the reflex is still there.

If September 11 happen today, we'd be arguing about it by dinner time. You know, whose fault was it? Which network do we trust? Which video was real?

Within a week, a conspiracy theory would have more views than the president's address. And within a month it would probably be a wedge issue. I hope that I'm wrong. I fear that I'm not. And technology is only half of it. The other half is what those 25 years contained.

We built a 24-hour news cycle that needed to be fed, and the world obliged. Iraq, Katrina, the financial crisis, the first black president, the night we got Bin Laden, Sandy Hook, Brexit, and Trump, in the same year a pandemic that killed more than a million Americans, George Floyd, January 6th, the fall of Kabul, 20 years after we went in, Ukraine, October 7th. Each arrived as the most important story of our lives, and each was displaced by the next before we had finished absorbing it.

A news cycle that fast doesn't let anything stay momentous. There's always something else. You want evidence? A few blocks from ground zero on Greenwich Street, there used to be a second museum, the 9/11 Tribute Museum, founded by the families.

The tour guides were survivors. They were firefighters. They were the mothers and the fathers of the dead. Over 16 years, it welcomed 5 million visitors from 141 countries. And then the visitors stopped coming.

More than 500,000 attended in 2011, 150,000 in 2019, 26,000 in 2021. It closed for good in August of 2022, three weeks before the 21st anniversary. Not because of a scandal or a lease dispute, but because a museum built on the promise that we would never forget could not stay open on the number of us who remember to show up.

And here's the part that should worry all of us. Anyone under 25 today was born after the attacks. That's a third of the country. They know 9/11 the way that I knew Pearl Harbor as a chapter, not a day.

Here's an example. The only Rudy Giuliani that they know is the post 9/11 Rudy. But there was another Rudy, a Rudy who provided the country with real leadership when we needed it the most as I told him this week.

(BEGIN VIDEO CLIP)

SMERCONISH: Now, I want to address all of those who will listen and all of those who will watch, and they'll say, you know, you spent a half hour with Mayor Giuliani and you didn't ask him about Trump, and you didn't ask him about this, and you didn't ask about that.

And here's what I want to say. I want to say, Mayor Giuliani, in the last 25 years, has pissed me off too. But I -- but I will be forever grateful for the leadership that he showed at a time when we needed it most in the aftermath of September 11th. I'd rather say that with you here than say it -- than --

RUDY GIULIANI (R), FORMER MAYOR OF NEW YORK: Thank you, Michael. I appreciate that. Of course, I've disagreed with you too in all that time. And I -- and I do believe I was doing the right thing during that period. It's the same Rudy Giuliani that was 9/11 as the Rudy Giuliani who stood up for a president who was persecuted. But -- and was persecuted himself.

But having said that, I understand that too. I understand that people disagree with me on things and never have held that a grudge or anything like that.

(END VIDEO CLIP)

SMERCONISH: If we don't restore some common experience, then God forbid there's a repeat of that morning, we'll not be able to do what we did in 2001. We'll not be able to come together. 9/11 showed us some of the worst of humanity. It also showed us some of the best. And both are reasons we should never forget.

You still have time to vote on today's poll question at Smerconish.com. Could AI wipe out humanity? Please sign up for my newsletter while you're there voting. You'll get the work of illustrators like Rob Rogers.

(COMMERCIAL BREAK) [09:53:59]

SMERCONISH: OK. There's the poll result so far, pretty decisive and a lot of votes. A tick under 30,000. Could AI wipe out humanity? Seventy-six percent of us, I'm in that affirmative vote, say yes.

Here are some of your social media reaction to today's program. Appreciate you following me in all the usual places.

Do you truly believe AI can kill humans? Today, you have officially become a conspiracy theorist.

Hey, Gurby, did you not hear the part where I said that the guy named Hubinger, who is at Anthropic, not he who quit, but the guy who's still there and is in charge of controlling AI, said that in his opinion, there's a greater than 10 percent likelihood in the next decade? Did you not hear that part?

So, it's not me being an alarmist. I'm not a conspiracy theorist. But in this case -- I'll tell you how I regard this. And it's funny to have had Richard Clarke here on the program today.

Yesterday, of course, you know, we honor those that we lost on September 11. Well, you remember a month in advance of September 11, there was the PDB, the President's Daily Briefing, Bin Laden determined to strike in the U.S.

[09:55:04]

That was the headline. I view this as the equivalent. We've just been given the PDB by somebody on the inside. Now, what are we going do about it?

More social media. Go ahead. What do we have?

I'm shocked. The commie from @CNN doesn't want election security.

That must be directed to somebody else. Next, what do we have?

Never forget. The right mayor at the right --

Yes. Jeff, I agree with you. Look, you know, people are like, oh, how could you be so nice to Rudy? Because I remember Rudy.

I remember Rudy on that day and at that time, and what we so desperately needed. And 25 years is a time of reflection. And I'm not going to forget that. And I'm not afraid to say it.

You heard me say to him. I said to him, Mayor, you've pissed me off, you know, in the last 25 years. He said, well, you did the same to me.

But I'd rather have the civil dialog with him. And I'll never forget what he did for the country. That's it. By the way, does that make me a commie?

If you missed any of today's program, you can always listen anywhere you get your podcasts. We thank you for watching and we will see you next week.

(COMMERCIAL BREAK)