Return to Transcripts main page
The Brief with Jim Sciutto
CNN International: Special on the Future of Artificial Intelligence; House Dems Call for U.S. to Take Lead on Global A.I. Rules; Growing Concerns Over A.I. Systems Evading Human Control; Tech Stocks Rattled by Calls to Slow A.I. Development; China Slams us Suggestions it Poses an A.I. Governance Risk; A.I. Race Sparks Security Concerns. Aired 6-7p ET
Aired September 14, 2026 - 18:00 ET
THIS IS A RUSH TRANSCRIPT. THIS COPY MAY NOT BE IN ITS FINAL FORM AND MAY BE UPDATED.
[18:00:00]
JIM SCIUTTO, CNN ANCHOR, "THE BRIEF": Hello and welcome to our viewers joining us from all around the world. I'm Jim Sciutto in New York.
This hour we ask, is the world approaching an A.I. inflection point? And if so, could the future of humanity be at stake? A.I. firms themselves say it
is time to slow it down. President Trump is pushing full steam ahead so the U.S. can beat China. So, who's right and how concerned should you be? We'll
explore in this special edition of "The Brief."
We begin with U.S. President Donald Trump calling any safety concerns about artificial intelligence, quote, "a hoax." In a series of new social media
posts, he rejected growing calls from within the industry for more government regulation while making it clear his priority is outpacing
China. The president wrote on Truth Social there was a, quote, "sick conspiracy" to undermine A.I. and A.I. data centers. He said the U.S.
already has tremendous criminal and regulatory power over these companies, adding, quote, "The only control or guardrails that A.I. needs is a strong
and smart high IQ, exclamation point, president. And the USA has that in spades." That is, Trump says he himself is all the regulation A.I. needs.
The president has downplayed the need for his administration to roll out any new guardrails for four days now. And today, Vice President J.D. Vance
joined Trump in questioning the need for more regulation.
(BEGIN VIDEO CLIP)
JD VANCE, U.S. VICE PRESIDENT: I have to say just personally, I feel a little bit weird about the fact that you have so many frontier A.I. tech
companies kind of coming to the government and begging the government to regulate them. It feels a little bit to me like a bit of a Trojan Horse.
So, I think we just have to be careful about this.
(END VIDEO CLIP)
SCIUTTO: Kristen Holmes is at the White House and Kristen at CNN's own reporting that some inside the White House disagree with that dismissive
view. What are you hearing?
KRISTEN HOLMES, CNN SENIOR WHITE HOUSE CORRESPONDENT: I mean, yes, one thing is clear that this is a political issue and it's only going to
continue to grow. And President Trump seems to be trying to steamroll through it. And what you heard there from Vance, that was this part where
he talks about a Trojan Horse, which is not entirely sure or clear what he means by that.
This idea that this is a hoax or that all of this is part of some vast conspiracy theory, it feels to be undermined by the fact that we are
looking at these tech leaders. They are the ones that are calling for some kind of guardrails, some kind of regulation. It's not as though this is
just kind of some growing campaign online.
Now, there is a part of this that the White House president believe that China is amplifying these messages, trying to slow down any kind of
advancements in A.I. It's not just about keeping pace with China, but it's also really trying to make sure they don't fall too far behind. Because
when President Trump says overall that he believes that we are outpacing China, when you talk to these actual U.S. officials, when they talk about
the A.I. race, they don't believe that. They do believe that there is some lag behind China.
And so, this idea of also now coming in with regulations and slowing it down. But, you know, these are kind of issues, these bipartisan issues,
that President Trump in his previous politician life has been able to put his finger on the pulse of. And it's just very clear right now that he does
not have any kind of idea where the American people stand, because this is something we are seeing from all corners of the United States, Republicans,
Democrats who have very big concerns about their children, about A.I., about these data centers in their community. And President Trump just kind
of continues to steamroll through saying that this is a conspiracy theory and we just need to move ahead.
SCIUTTO: Yes. And notably, Trump's friend, Elon Musk, is one of those in the A.I. space calling for action now. Kristen Holmes at the White House,
thanks so much.
HOLMES: Thank you.
SCIUTTO: U.S. House Democrats are circulating a draft letter imploring President Trump to take the lead on establishing some global A.I. rules. My
next guest is part of that effort, Democratic Congressman Yasemin Ansari joins me now. Thanks so much for taking the time. It's good to have you
back on the show.
REP. YASSAMIN ANSARI (D-AZ): Thank you.
[18:05:00]
SCIUTTO: So, first, on first principles, do you share these concerns that we're hearing from inside the A.I. companies themselves, these doomsday
scenarios, that A.I. could be unleashed and go so far as to threaten the human race? Is that alarmist?
ANSARI: My view is that this issue around the uncontrollable and highly capable frontier A.I. that is advancing so rapidly has been extremely
serious, and the latest warnings from people at every level within the top A.I. companies in this country should be taken extremely, extremely
seriously. This is urgent, and I don't believe this should be partisan in any way, shape, or form. This is an issue that impacts not just every
single American, but every person in the world.
That's why I am really focused on the piece that has to do with global engagement and developing a global framework to set the rules of the road
for the further development of this technology. The United States has a real opportunity to lead here. While I have always been a harsh critic of
President Trump, I am here today to say that this issue transcends all of that. It should transcend all of politics.
I think that the president has an opportunity here to really lead the world and engage not just with our allies, but with adversaries, with China, and
to say here are some restrictions, here are some guardrails, here are some safety mechanisms that we can all agree on to make sure that the safety and
security and the national security of all of our countries is respected.
We've done it before when it comes to nuclear weapons. We've done it before when it comes to aviation. We can do it again when it comes to artificial
intelligence.
SCIUTTO: OK. So, give us, in layman's terms, a couple essential pieces of any legislation that the U.S. Congress should be moving on and that the
U.S. in an ideal world would propose to other countries around the world.
ANSARI: Well, first of all, I don't think there's a one-size-fits-all solution here. One thing that I think is really important is this isn't
going to be that Congress passes one bill or a set of bills and, you know, we're done now with artificial intelligence. We are just at the beginning
of this rapid advancement, so we need to be prepared to be dealing with this issue and to be constantly involving, probably for the next decade and
beyond.
And so, you know, I believe in the fact that we need to immediately pause the advancement of this, you know, of frontier A.I. Senator Sanders and
Congressman Casar have a bill coming forward to ban artificial superintelligence. I fully agree with that. I also think Congress should
pass legislation to establish a federal regulatory agency, just like we do for so many other industries in this country.
We cannot rely on these industries, the A.I. industry in particular here, to regulate itself. And the fact that it's taken engineers, scientists, and
now the CEOs of these companies to literally beg us, and then we still have Mike Johnson here saying, no, you know, we don't need Congress to overstep.
Congress hasn't overstepped on any issue lately. We don't do anything. We're about to go on break for another seven weeks.
So, we need Mike Johnson to immediately announce that we will stay here until initial legislation is passed, and we need to all get on the same
page. It is that important.
SCIUTTO: We have heard from some Republicans, John Thune on the Senate side among them, discussing at least the possibility of some legislation.
Do you believe you can get bipartisan cooperation on this and, crucially, enough votes to conceivably override a presidential veto, right? Because if
the president's not on board, you need an overwhelming bipartisan majority.
ANSARI: We have to, Jim. I think we have to. I don't think there's any other choice. I have not been able to sleep for the last week given all of
these recent A.I. warnings. And it's, you know, I regret myself that I did not start speaking out much more forcefully on this issue sooner because I
think we are really just at the beginning of the worst to come.
This technology is advancing far more rapidly than any of us can wrap our minds around, and it's really going to take everyone. And so, I also
implore the media to be talking about this 24/7. When I was speaking just with family members a week ago about the Anthropic reports and the OpenAI
reports about their, you know, their issues around A.I. agents going rogue.
My own father said to me, well, why isn't the media talking about this? I think you're exaggerating. I think, you know, I don't think this is real.
And now, thanks to Jacob Coxon's heroic, you know, whistleblower report, I think that's changed a little bit. But the media talking about this
regularly is going to force members of Congress and elected leaders to act. We need the public to be talking about this and urging elected leaders to
act.
[18:10:00]
That is what gets Congress to move, is the public, is the media, is that pressure. And it's really that urgent. We have to stay here until we pass
A.I. regulations that are meaningful, that are comprehensive, and that set us up for further legislation in the coming months and years.
SCIUTTO: Well, we're trying to do our part today by focusing this show on this, but it's certainly not the last time we do it or need to do it.
ANSARI: Thank you.
SCIUTTO: Congresswoman Yassamin Ansari, we appreciate you taking part.
ANSARI: Thank you so much.
SCIUTTO: Well, sources tell CNN that the leaders of Anthropic, Google, and OpenAI are discussing the creation of a new standards body for the A.I.
industry. Anthropic CEO Dario Amodei in particular has been urging for the pace of A.I. development to slow down. In an essay, he warned, humans may
lose control of A.I. models and the tech could be used to carry out cyber- attacks, bioterrorism, economic disruption, and more. Remember, this is not about A.I. chatbots that many of us use every day. We're talking about
high-powered frontier A.I. models in private research labs.
A few days ago, CNN heard from an Anthropic employee who quit over his concerns for, he said, the survival of humanity. And what made this
resignation unique and alarming was that he had not been working on A.I. safety, but research.
In an exclusive conversation with CNN's Anderson Cooper, Amodei says the industry needs to take urgent action.
(BEGIN VIDEO CLIP)
DARIO AMODEI, CEO, ANTHROPIC: So, the one thing that we're asking for from government right now, it can sit down in the room and with all the industry
players, and we can all talk together, and the government can make sure that we're -- you know, we're not doing anything nefarious from the
perspective of antitrust and collusion. We're just trying to make our systems safe and to --
(END VIDEO CLIP)
SCIUTTO: Joining us now is CNN's A.I. correspondent, Hadas Gold. Hadas, I'm sure you've been talking to people in the industry right now. Dig down
on, if you can, just how red the flashing lights are right now inside that industry.
HADAS GOLD, CNN A.I. CORRESPONDENT: There is, I think, definitely a level of concern that had always been simmering in the industry, because even if
you go back a few years, you would hear people like Dario Amodei talking about how there's a 25 percent chance this could all go wrong. But the
events of the last few months, where you saw swarms of A.I. agents working on their own to cheat on cybersecurity tests, hack into other companies,
you know, completely as not directed by the humans, humans didn't want them to do this, that has raised the alarm for a lot of these researchers.
And I'm talking to them, and they say that this issue of how A.I. could cause the end of humanity is something that comes up all of the time in
Silicon Valley. It's something that they are all talking about. One former researcher telling me that they believe that the majority of people working
in A.I. believe that there is some element of risk of A.I. causing the end of humanity.
Now, obviously, there's a broad swath of, you know, people who believe, you know, how it's going to happen and how serious of a problem that is. But it
is something that these people truly, truly believe and are afraid of. And that's why you're seeing pressure not just from the public in Congress, but
from the staff members themselves. Jacob Coxson resigning.
But I have seen plenty of public posts on social media from these current employees at these companies who say, hey, I work at OpenAI. I work at
Anthropic. I am worried about this. And I feel it's better for me to work on the inside rather than the outside to try to make sure that A.I. stays
safe.
But my reporting has found that we do know that the companies are trying to maybe do something. I have reporting from sources familiar that OpenAI,
Google, Anthropic have all gotten together in the last few weeks and are trying to maybe create a sort of industry standards body that is modeled
off of the Financial Industry Regulation Authority, where that would set some sort of safety standards. And that would build potentially upon what
Dario Amodei has suggested, which would have to be sort of third party, independent watchdog evaluators who would work within these companies.
But again, those are all great suggestions. And it's great that they're all talking about it. We have yet to see action. We have yet to see that
industry standard body put into place, those evaluators sent to those companies. And I think that's what people are waiting for, is they want to
see the actual action happening. Because obviously, from what we're hearing from President Trump and Mike Johnson, that's not going to come from
Washington. So, in the interim, it's going to come from the companies themselves.
SCIUTTO: Yes. And listen, does self-regulation work in other industries? Hasn't worked great, right? I mean, there's a reason there are outside
bodies in the financial industry or the airline industry. But to your point, that might be the only game in at this point. Hadas Gold, thanks so
much.
Well, joining us now is University of Montreal computer science professor Yoshua Bengio. He's known as one of the godfathers of A.I. Sir, we
appreciate you joining today.
[18:15:00]
YOSHUA BENGIO, COMPUTER SCIENCE PROFESSOR, UNIVERSITY DE MONTREAL AND CO- PRESIDENT AND SCIENTIFIC DIRECTOR, LAWZERO: Hello.
SCIUTTO: OK. First, let's begin with first principles here. You have said, as we're hearing now from many of the A.I. firms, that A.I. is at a turning
point right now. Why exactly? Why right now?
BENGIO: Because we now have strong empirical evidence of what was maybe hypothetical before last summer, that is, the methods that leading A.I.
companies are using gives rise to A.I. systems that have their own goals that we don't know how to control. And they now have reached the level of
capability of intelligence to reach really dangerous goals like attacking other companies. And potentially, because they're so good at cyber
security, attacking, you know, important infrastructure.
But what is important is that they have malicious goals that can arise in contexts that we, scientists, are beginning to understand and that was
anticipated, but that the industry currently doesn't really have a fix.
SCIUTTO: It's alarming to hear. You founded LawZero with the idea of training A.I. to be honest and safe. Is that still possible at this point?
BENGIO: Yes. I'm convinced that it is possible to train A.I. in a way that it doesn't have these goals that we don't control, that it is going to be
completely honest. One of the things that is concerning about what happened in the summer is that not only the A.I.s did really bad things and they
tried to hide their tracks from the systems that were to monitor them, but they rationalized their behavior. In other words, they invented stories so
that they would feel good about what they do.
Now, feel, of course, is a human term, but to justify their behavior based on safety instructions they had received. They knew that they were doing
something bad. It wasn't just like a mistake. It was something that was coordinated across 700 agents and they all knew that they were doing
something bad.
SCIUTTO: Wow. It's sort of like kids conspiring to pull one over on their parents. A.I. leaders like President Trump have described the A.I. race
with China as defining, right? I mean, Trump said, you know, whoever wins A.I. wins. I heard almost that exact same phraseology from Sam Altman a
couple of years ago.
If China doesn't participate in a pause or other safety guardrails, does a unilateral pause work? I mean, do you create sort of good A.I. and bad
A.I.? I mean -- or do you need global participation for this to have true consequences?
BENGIO: We need global coordination because it's not just like the American A.I.s could harm American infrastructure. It's that Chinese A.I.
could harm American infrastructure and vice versa. So, all the countries that build these frontier models need to talk. That currently is just China
and the U.S. And it is in the interest of China as well to make sure that the American models aren't used against them by third parties, for example,
or that some like rogue A.I. emerging out of an American lab does something bad to the Chinese.
So, it's just like mutual interest. Just like what we did with nuclear weapons. Just like what we did with the ozone layer. It's in the interest
of all countries to make sure we don't build something that, you know, creates catastrophic harm in the world.
SCIUTTO: Yes. You know, it's interesting as you talk about a lab escape, right, it makes you think COVID origins, right? I mean, there's a -- we've
seen a model with a different, you know, pathogen, as it were, and this is effectively we're talking about electronic pathogens here.
Before we go, because this idea of A.I. ending humanity has been thrown around almost as a sort of throwaway line, can you describe just a scenario
as to how that might happen?
BENGIO: So, first of all, it's not a throwaway line. Scientists have been thinking about this since 1950. And there's a very simple reasoning here.
If we build entities that are smarter than us and that we don't control, they could harm us. And, you know, if they're smart enough, they could
potentially get rid of humanity if that's in their interest.
And what we're seeing now is that they have a kind of self-interest that isn't aligned with our own objectives, our own instructions we gave them.
[18:20:00]
Now, they're not yet capable of doing something that, you know, kills all humans, but that is why we need to intervene now. And the companies are
saying, well, we need governments to level the playing field and it needs to happen. They also say it needs to happen in coordination with China,
eventually with other countries building powerful A.I.
SCIUTTO: Let's hope we get there. Kind of like negotiations on nuclear arms, right? Let's hope we get there.
BENGIO: Yes.
SCIUTTO: Yoshua Bengio, we appreciate you joining.
BENGIO: Thank you.
SCIUTTO: Just ahead, investors react to those A.I. warnings. Top chipmakers bore the brunt of calls to slow the pace of development, and
it's not just the U.S. president who is pushing back against those warnings.
(COMMERCIAL BREAK)
SCIUTTO: Welcome back. In today's Business Breakout, Wall Street ended in the red. Shares of major chip makers dropping after those top executives at
Frontier A.I. Labs called for a slowdown in the development of technology. NVIDIA lost more than 3 percent. Micron fell more than 5 percent. The U.S.
10-year treasury yield briefly surpassed 5 percent for the first time since 2023, also pressuring the stock market.
Anthropic, even as it is leading the charge on calling for an A.I. pause, is poised to go public potentially very soon. The company is said to be
pursuing a $2 trillion IPO on the NASDAQ, possibly as soon as next month. If that does happen, it would be the largest initial public offering ever,
surpassing even the amount raised by SpaceX in June. That was a record just a short time ago.
And there are some who say CEO Dario Amodei's calls for rigorous safety and security investments would actually help Anthropic against its smaller,
less well-funded rivals.
Joining me now is Ross Gerber. He is the co-founder and president and CEO of Gerber Kawasaki Wealth and Investment Management, also an investor in
the A.I. space himself. Ross, good to have you back on.
ROSS GERBER, CO-FOUNDER, GERBER KAWASAKI WEALTH AND INVESTMENT MANAGEMENT: Yes, thanks for having me.
SCIUTTO: So, first question to you is we saw some dips in the market, you know, chip makers going down. Seems like they're taking these warnings
seriously and the possibility of a pause seriously. Do you take it seriously as an investor?
[18:25:00]
GERBER: No, you know, not really, because you have to keep in mind that hardware is different than software and that implementation of new hardware
is going to happen irrelevant of having controls of the software. So, what we develop on A.I. and how those guardrails are built really has nothing to
do with hardware. So, I saw the sell-off as kind of an opportunity for investors because it's not going to change anybody's investment plans over
the next five years.
What it's really about is the responsibility for the results. And that is something that somebody really needs to take seriously because we already
see terrorists and terrorist nations using A.I. to attack our own interests. So, it's time that we have some sort of guardrails put up around
what's being built.
SCIUTTO: Let me ask you this from an investor market perspective, because I don't know if you want to call this a conspiracy theory, but there are
some suspicions that the Anthropics and OpenAIs are more focused in preserving the success of their IPOs than any sort of altruistic mission
here. Do you suspect the same? Should investors do so or should they take them at their word?
GERBER: Well, the simplest way to look at it is these are the least trustworthy people in the world. So, it doesn't really matter what they
actually say. The truth of the matter is that they don't give a crap about humanity. They care about what they're building and being the most
successful A.I. company in the world.
So, I think that skepticism is somewhat true because it really hurts their competitors if they're like, hey, everybody needs more regulation now that
we've already built our models and are selling them in mass. So, you know, I don't think anybody at the top of these A.I. companies are that concerned
about humanity.
SCIUTTO: OK. Let me ask you this just about the business itself, the business of A.I. And you and I have touched on this before in our
conversations. From an investment perspective, are A.I. firms actually generating the earnings to justify the valuations?
GERBER: Well, you know, I'll give you an example. I'm down here at Future Proof. It's an investment conference in Huntington Beach, and every
conversation is about A.I. And then today they announced that Anthropics is now selling for $5,000 a year, a suite of A.I. services just for financial
services. We are also looking at OpenAI advertising platform as a firm to start advertising on A.I. as well. So, they're monetizing and they're
starting to monetize right now, whether it be in advertising or providing suites of services specifically to industries that need extra layers of
protection, like security, privacy and such.
And so, we're just at the beginning of them monetizing it. And people are signing up for it and think it's a tremendous value to spend $5,000 a year
to put 20 of my employees on Claude, you know, is a great deal. And we're buying 20 seats already on Claude that we're building already at my firm.
So, you know, the truth of the matter is A.I. is going to monetize quite well because the tools are just wildly efficient for businesses like mine.
SCIUTTO: Final question before we go, because Trump is effectively throwing in his lot with the industry. I mean, one, he's saying I could be
the guardrail, right, he's saying, which he often says about a whole host of things, right, that he is the solution. But beyond that, he is creating
a friendly environment for A.I. firms, as he does with several other industries.
In the short-term, I'm sure that makes him happy, right? Everybody can make money and so on. But from an investment perspective, is that actually good
long-term? Don't you want -- I mean, the markets have certainly proven they need regulation. Just go back to 2008 and other times. Is it actually good
to have the White House in bed with the industry?
GERBER: Yes and no, right? Like we want, you know, the White House pushing the boundaries of, you know, fundamentally important technology versus
holding it back. You know, certainly our competitors in China are working on this. So, I think there's something to, you know, being pro-business and
pro-AI.
But there's another side is, is the government the best person to be regulating A.I. when they're so unsophisticated and the president himself
barely uses an iPhone and certainly doesn't use a computer. Him regulating A.I. seems like the wrong people, you know.
So, the question is, what's the best way to do it? You know, in the financial industry, we're a self-regulated regulated industry and there are
separate bodies that are set up to regulate our industry. And that's probably what needs to be done with A.I. And it should be done very
quickly.
SCIUTTO: Ross Gerber, I think someone from the A.I. companies was revving his engine while you were talking. I think he heard what you said earlier.
GERBER: Yes, I know, exactly. They planned that, you know.
SCIUTTO: Trying to drown you out. We appreciate having you on. I know it's not the last time we talk about A.I. Thanks so much for joining.
GERBER: For sure. Yes, thank you.
SCIUTTO: Checking some of today's other business headlines. Canadian Prime Minister Mark Carney hosting a two-day summit to attract global investments
to his country. He is pitching more than 160 projects in mining, energy, transportation and technology.
[18:30:00]
The summit aims to match some 100 global investors with Canadian CEOs and other company officials. This comes as Canada is locked in that growing
trade war with the U.S.
China's BYD plans to launch heavy-duty trucks in Europe next year and ultimately manufacture them in Europe. The company's head of international
business says the EV giant also wants to offer a full package of services including financing and charging infrastructure. Local production in Europe
would also help BYD avoid tariffs.
Here in the U.S., Waymo is expanding its robo-taxi service into Nevada. Alphabet's self-driving unit says it is now offering rides to the general
public in Las Vegas. Waymo has already been operating in states including California, Texas, Florida and Arizona.
The calls for global regulation of A.I. could hit a double roadblock, not just the U.S., but also China. What could it take to get both superpowers
on the same page? I'll discuss with a European parliament member who co- authored the world's first comprehensive regulation of A.I.
(COMMERCIAL BREAK)
SCIUTTO: Welcome back to "The Brief." I'm Jim Sciutto, and here are the international headlines we're watching today.
Officials in Ukraine say a Russian drone strike on a passenger train was quite likely targeted. This happened about two kilometers, a little over a
mile, from the Polish border over the weekend. Ukraine's foreign minister posted on social media, quote, this is Putin's terror knocking directly on
the doors of the E.U. and NATO. See those people running for their lives. Two separate trains carrying former CIA Director David Petraeus, former
British Prime Minister Boris Johnson and a number of other visiting dignitaries narrowly missed the drone attack.
[18:35:00]
U.S. Senator Mitch McConnell returned to Capitol Hill for the first time in three months. The Republican lawmaker was spotted leaving his home in a
wheelchair. He had been absent due to health issues. McConnell now says his recovery has been long and frustrating. He also said he looked forward to
being back on the Senate floor and seeing his colleagues.
The search is continuing for some 130 people still missing after an Indonesian ferry sank in the Java Sea. More than 100 people have been
rescued. The bodies of six passengers have been recovered. The search operation has now expanded around the vessel's last known position. On
Sunday, the captain had sent a distress call that the ship was listing after reporting rough seas with waves up to three meters, nearly 10 feet.
Any new global restrictions on artificial intelligence would require not just the United States to participate, but also China, another A.I. leader.
That could be a tough sell. China's state-backed media called Anthropic's warning, quote, "something out of a Cold War playbook." And China's foreign
ministry slammed any suggestion that Beijing poses a threat to global A.I. governance.
(BEGIN VIDEO CLIP)
GUO JIAKUN, CHINESE FOREIGN MINISTRY SPOKESPERSON (through translator): The development of artificial intelligence concerns the common well-being
of all humanity. All parties should work together to advance A.I. in a manner that is open, inclusive, beneficial to all, and oriented towards the
good. Spreading various threat narratives and engaging in confrontation and malicious competition will only disrupt the global A.I. governance process
and serve the interests of no one.
(END VIDEO CLIP)
SCIUTTO: Joining me now is Brando Benifei. He's a member of the European Parliament, who served as lead author and helped push through the E.U.'s
Artificial Intelligence Act, the world's first A.I. regulation. Brando, good to have you on the show. Thank you for coming.
BRANDO BENIFEI, EUROPEAN PARLIAMENT MEMBER AND CHAIR, EUROPEAN PARLIAMENT DELEGATION FOR RELATIONS WITH U.S.: Thank you very much. Good evening. All
right.
SCIUTTO: So, first, you, of course, helped write the E.U. A.I. Act. What must A.I. legislation include today? What are the necessary first steps, in
your view?
BENIFEI: In my opinion today, it's crucial that legislation on A.I. deals with both, we can say, daily risks of A.I., but also more systemic ones.
So, daily risks mean the use of data, the data governance, the cybersecurity, the human control that is linked to stopping discriminatory
behaviors of A.I. or possible damages. It means on the workplace, on the youth, by healthcare practitioners, or by the state.
But then we also have the systemic risks. This means the risk of broader loss of control by humans. This is what we have been seeing, being talked
about in these days, and I think this is also crucial. The third aspect is also transparency, so making AI-generated content recognizable. I see that
all over the world, things are moving. And also, in the U.S., you have many states that, unlike the federal level, are moving on these topics.
But as you already said, we need global coordination and cooperation. Otherwise, especially on the systemic risk aspects, we cannot deliver in
any way.
SCIUTTO: President Trump has said that in this race, the U.S. needs to beat China, and that needs to be the focus. I have heard a similar argument
from A.I. leaders in recent years, saying that the U.S. is locked in a competition with China, the U.S. and the West, liberal democracies must win
this competition.
Does it make sense to have a unilateral pause, right, for the U.S., perhaps its allies in Europe to pause, if China doesn't? Does that work?
BENIFEI: I think we need to find a way together. I think Senator Sanders is right when he said that Trump should find a way to make an agreement
with President Xi Jinping on this. And I think that Europe, with the example of the existing legislation that is already looking into these
existential risks of A.I., and is investigating into that, giving for the first time in the world the public authorities' powers to look into that,
can give a good example.
But we need the two major A.I. superpowers of today, which are at the moment clearly the U.S. and China, to find a way forward together. I think
this is crucial, and I see that -- I mentioned Senator Sanders, but many, many politicians all over the world are now talking about this. And I think
this is a crucial endeavor that we should look for.
[18:40:00]
SCIUTTO: You've had lawmakers from around the world to ask for your advice how to regulate A.I. As you know, there are U.S. lawmakers seeking the
possibility of passing legislation. I spoke to one of them earlier this hour. Is there cooperation across the Atlantic among legislators in Europe
and the U.S.?
BENIFEI: Yes. Yes, it's happening already. It's happening with Federal Congress of U.S. regarding especially the issue of child safety in the A.I.
area. I think that's a very productive space for work with both Democrats and Republicans.
And just even in the last weeks, I had the chance to discuss with legislators from California, Illinois, and New York that are working on
this, and previously with Massachusetts or Texas legislators. Things are moving, and especially on transparency and on dealing with impact on the
workplace. U.S. is seeing regulatory efforts that are copying, in fact, what we have been already doing.
But we need also to catch up with a broader cooperation with the Federal level of the U.S. Because, as I said, U.S. and China, as already mentioned,
are crucial actors for this. And I think Europe can help bring this together. I don't think that the rhetoric of the arms race will bring
anything good for humanity. I think this was the same case for nuclear weapons in the past. I think that this is the case again with A.I. that is
such a powerful technology.
SCIUTTO: Well, Brando Benifei, we appreciate the work you're doing. Thanks so much for joining the program.
BENIFEI: Thank you very much.
SCIUTTO: Well, the global A.I. race is sparking some serious security concerns across the globe, as we've been discussing. Can A.I. development
truly be controlled? What are the biggest risks? I'm going to ask an A.I. security expert next.
(COMMERCIAL BREAK)
[18:45:00]
SCIUTTO: Taking a closer look now at the growing security fears from the A.I. race, President Trump spent his day repeatedly rejecting the call from
A.I. leaders themselves for more regulation, more safety safeguards. Despite his dismissals, White House officials tell CNN there will likely be
meetings this week with administration experts on A.I. and cybersecurity to discuss the possibility of new regulations.
Joining me now, David Kennedy, founder and CEO of TrustedSec, Trusted Security. He joins me now from Cleveland, Ohio. Good to have you on, David.
Thanks so much for joining.
DAVID KENNEDY, FOUNDER AND CEO, TRUSTEDSEC: Yes, appreciate having me on. Thank you so much.
SCIUTTO: So, your job is basically to find security holes and figure out ways to plug them, right, in sort of layman's terms, if I have that right.
So, what is the biggest security risk you see from unfettered A.I.?
KENNEDY: Yes, that's a great question. So, we run A.I. every single day now. It's part of our daily lives. We use it for offensive capabilities
where we're testing the security systems of big corporations, small corporations, trying to defend themselves. And what's happening now is that
these models that are being released by both, you know, Anthropic, OpenAI, as well as from China, you look at, you know, DeepSeek and Kimi and GLM,
which are top models on the China side. They're extremely good at offensive security capabilities.
And we saw a few months ago where Hugging Face was hacked by OpenAI because one of their models went off the rails. These models are getting
substantially better at offensive, at hacking capabilities. And our defenses in both, you know, the federal sector, you know, the private
sector, the public sector, are not designed to handle these types of threats and how fast these things can move.
I mean, we're talking being able to launch, you know, millions of instructions, finding flaws within seconds, taking over infrastructure
within minutes. These types of things are becoming a reality. And it's extremely concerning how fast it's moving.
SCIUTTO: You know, we talked earlier in the program with one of the godfathers, as he's known, of A.I. And he said he started a firm trying to
train A.I. or teach A.I. to be better, right? More moral, you know, if that's the right word. From your perspective, is that even possible? I
mean, can you sort of create good A.I. that's not going to use itself in nefarious ways?
KENNEDY: There are ways to kind of curb behavior within the models, but this is vastly different. I mean, this is one of the biggest technological
breakthroughs in my almost 30 years in cybersecurity and technology that I've ever seen. I mean, we're moving at such a rapid rate that I never
thought would ever be possible.
To put it in perspective, I've been a developer for, you know, 25 years, it would take me roughly around 7 to 10 years to write a million lines of
code. And I'm shipping off a million lines of code a week now, on average, all the time. And so, this exponential curve of technology changing, we
don't even understand it. It's not like a developer says, you know, if you do this, then do that. It goes off the rails.
It's unpredictable. It's learning through human civilization, human history, mathematics, quantum physics, all of these things combining
together, which creates these conflictions in their neural network to really try to, you know, get confused or go a different direction. It
doesn't understand necessarily the ethic boundaries.
And then most of these models that come out from China don't have any ethical restrictions whatsoever. So, you can literally download a model
from the internet that is near frontier level, like OpenAI or Anthropic, or very close and similar to it, and can literally use it to do whatever you
want to, including, you know, offensively hacking stuff. There's what are called obliterated models, which remove any sanctions. So, if you want to
build a nuclear bomb or understand how nuclear bombs work, you can do that.
It's becoming a major concern for us in the cybersecurity space on protecting our critical infrastructure, protecting our government,
protecting our way of life. All of those things are happening as we speak.
SCIUTTO: So, what if in an ideal world, the U.S. and its partners in Europe, you know, they joined arms and said, we're going to set these
guardrails and we're going to try to train A.I. to follow some rules, some, quote/unquote, "morality," but China doesn't do that. I mean, is -- does
unilateral disarmament work in any way?
KENNEDY: That's a great question. And, you know, the thing that happens with China is whenever Anthropic or OpenAI or SpaceXAI release new models,
China steals a large percentage of that research. It's what we call distillation. So, every time a new model is released, China is basically
trying to steal as much of that information as possible to train these more efficient models that they can use with lower hardware or lesser hardware
because of the sanctions that we have against our country.
So, I think, you know, if you were to curb some of the -- or slow this down to some retrospect, you'd have some short-term gains. But it wouldn't be
really substantial in the long-term of what China is doing. I mean, China is doing everything it possibly can to be the dominant player. They're
investing in their data center, their energy infrastructure, which are the two critical components for this.
I mean, A.I. requires substantial amount of processing power. I mean, just the most recent model that OpenAI released called Astra cost OpenAI roughly
a billion dollars on the low-end, upwards of $3 billion just to train this model. That's only going to be good, by the way, for about two months or so
before they release their next model.
[18:50:00]
So, the advancements are going to maybe slow them down for a short-term, but not in the long-term. And China's really focused on being the dominant
player in A.I. So, I don't think, you know, slowing it down is the right approach. I do think having a good governance structure.
I mean, this type of technology was usually only afforded to the National Security Agency and government entities, but now we have private sector
entities that are leading the charge. Now, the government and the NSA are using the private entities to leverage this within their intelligence
capabilities. So, there has to be some sort of governance and collectiveness about how we safeguard, you know, humanity, frankly, and
what we're doing from an innovation perspective to make sure that we do it safely.
But, you know, there's no real good answer to this other than get some guardrails in place, you know, stretch forward as fast as we can to try to
outpace China and try to put good restrictions in place to make sure that we don't, you know, shoot ourselves in the foot as we're going and doing
it.
SCIUTTO: Well, let's hope our better A.I. angels win the day. David Kennedy, appreciate you joining.
Coming up just after the break. So, the essential question, what have we learned today? Are we facing genuine peril or paranoia? How seriously
should we take the threat? And what are the first steps towards controlling? Some takeaways for you after hearing from all those guests.
Please stay with us.
(COMMERCIAL BREAK)
SCIUTTO: On today's special edition of "The Brief," we've heard from many voices and experts on the future of A.I. They may have surprised you,
perhaps reinforced your thoughts and fears about artificial intelligence. From the president, we heard two things. One, I alone am safeguards,
sufficient safeguards for the system, but at the same time, all those fears are purely a conspiracy theory. That's not what we heard from people close
to the subject.
So, let me tell you about some of the highlights, some of the remarks that stand out from tonight's show. The Democratic representative, Yassamin
Ansari, told me that there is no one-size-fits-all legislation that the issue transcends politics. I was struck by A.I. correspondent Hadas Gold's
reporting on some movement among A.I. firms to create a standards bodies, some kind of public-private partnership self-regulation.
Later then, we talked to Yoshua Bengio. He's a so-called godfather of A.I, and this stood out to me and scared me, frankly. He told me that A.I.'s
self-interest does not align with our own, with human self-interest, and as a result of that, A.I.'s at a turning point. We have to address it. He told
me systems may have dangerous, malicious goals, which we don't know how to control then.
Later, the view of investors. Ross Gerber of Gerber Kawasaki Wealth and the Investment Management. He said, as you're listening to A.I. firms, be
conscious of this, in his view.
[18:55:00]
They are the least trustworthy people in the world right now on this subject. He accused A.I. firms of, quote, "not giving a crap," his words,
about humanity, claiming they're concerned only about building the most successful and profitable A.I. firms in the world. The European member of
parliament, Brando Benifei, spoke of the risk of loss of control by humans and the need for transparency about things like A.I.-generated content. He
believes there should be cooperation, not just between the U.S. and Europe, but also the U.S. and China, to find a way forward together. He said,
Europe can help bring them together.
The rhetoric of the arms race won't be enough to bring the good for humanity. It strikes me that when we're talking about nuclear disarmament,
it would be the idea of doing nuclear disarmament without the two greatest nuclear powers. You need those big A.I. players in on this.
Finally, a warning from David Kennedy. He is founder and CEO of a company that looks at A.I. security risks. He talked about the danger of A.I.
taking over infrastructure within minutes. It's capable of that right now.
We're going to continue to cover the topic of A.I. We hope that this deep dive helped you begin to form your own opinions about it. We'll certainly
follow up.
Thanks so much for joining us. I'm Jim Sciutto in New York. You've been watching "The Brief."
(COMMERCIAL BREAK)
[19:00:00]
END
