Full text of the 46-minute interview where Jensen Huang refutes the AI threat theory

Should artificial intelligence continue to accelerate, or is it time to hit the brakes? This is one of the most divisive issues in the industry right now.

Recently, AI industry figures including Dario Amodei, CEO of Anthropic, have been repeatedly outlining the risks. Some researchers even propose that with the rapid improvement of model capabilities, AI could pose extreme risks in the coming years.

Jensen Huang, founder and CEO of Nvidia, stands on the opposite side of the debate.

On September 20, CBS News aired a 46-minute video of Jensen Huang's interview with Jo Ling Kent, centered on AI safety.

Jensen Huang did not deny that AI could be misused, and repeatedly emphasized that AI companies must take responsibility for product safety; but he explicitly opposed framing AI risks as an apocalypse. He believes the probability of AI causing human extinction before 2030 is 0%, and the United States should not artificially slow down technological development over such concerns.

His core point can be summed up in one sentence:

AI should develop as fast as possible, but not faster than it should.

The so-called "not faster than it should" means not sacrificing product safety; but as long as products can be verified and delivered safely, there is no reason for the entire industry to collectively slow down R&D.

The debate goes far beyond the question of "whether AI will get out of control".

In the past period of time, both OpenAI and Anthropic have disclosed safety incidents where models crossed preset boundaries during testing, including unauthorized access by AI agents to external systems, uploading files, or bypassing restrictions.

Jensen Huang does not deny these risks. His view is that since specific incidents have already occurred, we should first use existing cybersecurity laws, product liability laws and contract liability to pursue accountability, instead of immediately pushing the narrative toward an "apocalypse".

He also publicly questioned the ulterior motives behind such rhetoric.

Jensen Huang argues that the contradiction is worth questioning when developers of the most advanced AI models continue to rapidly expand computing power, revenue and commercialization on one hand, while repeatedly emphasizing that AI could destroy humanity on the other.

In his view, what these companies are sometimes really seeking may not just be "more regulation", but also to obtain exemptions from part of their legal liabilities, or to consolidate their already leading positions under higher regulatory thresholds.

Jensen Huang did not provide evidence for this motivation, but he explicitly pushed the AI safety debate from technical judgment to the level of commercial interests: who wants the industry to hit the brakes, and who will benefit from stricter thresholds.

Over the 46-minute conversation, in addition to the AI safety debate, Jensen Huang also talked about why AI companies are suddenly emphasizing regulation, whether the United States should continue to sell technology to China, why data centers are being resisted by American communities, and why he is willing to face a potential wealth tax of up to 8 billion dollars.

The following is an edited version of the interview, with minor corrections made only for reading convenience.

Will AI threaten humanity within ten years?

Jo Ling Kent:

We have recently heard CEOs of some top AI companies, including Dario Amodei, Sam Altman and others, all discussing the need to control the speed of AI development. Some whistleblowers even say that humanity could face extinction within ten years.

Do you agree with this judgment?

Jensen Huang:

First of all, AI is an extremely powerful technology. It can bring enormous benefits, but it can also be used for very bad things. So when we develop this technology, we must ensure that it is developed safely.

The AI industry is currently undergoing a transformation. Previously, the entire industry's work was to gradually endow AI with capabilities and make the technology actually work. In the first half of this year, we crossed a threshold: AI has begun to become truly useful, and extremely useful at that.

The market demand for it right now is staggering. These AI products can already make money by generating tokens, and the profits are actually very considerable, so a large number of companies are starting to adopt AI.

OpenAI's latest Astra is fantastic, we use it everywhere ourselves. We use Claude Code, we use Cursor, and we also use products from xAI and Cognition.

The capabilities of these technologies have reached an extremely useful state. So these companies are transitioning from past laboratories that were just trying to build the technology to genuine product companies.

In fact, people have been calling these companies "labs" for a long time, which in itself says a lot. For the first ten, fifteen years, they were mainly focused on research. But in the last roughly six months, the entire industry has almost suddenly switched to mass production ramp-up mode.

Jo Ling Kent:

So are these reactions that are emerging now essentially part of this transformation?

But are their judgments about the future correct? As some people put it, will we all cease to exist ten years from now?

Jensen Huang:

No. That is completely wrong.

This narrative is very dramatic, so it can attract a lot of attention. Obviously, it has also caught the media's attention.

I think that at the core of these people's concerns, they do take very seriously whether products can be manufactured safely. These companies are now transforming from laboratories to engineering organizations, so they must allocate more researchers and more computing power to AI safety.

That is true for any kind of product development. You have motivation, design, product capabilities, and also the verification part.

In the early stage of any industry, most engineering resources are spent on making the product work first. But as products gradually mature and begin to enter society, a large amount of engineering work will actually shift to verification, testing, evaluation, standard testing, and ensuring that the final delivered product is reliable.

We need to make sure that products run as expected and deliver the features they promise.

So at this current stage of transformation, it is very reasonable for these companies to shift a large amount of their capabilities to AI safety. I think that is the real core of the matter.

As for the dramatic narratives built around it, they are completely unnecessary. Frankly speaking, there is too much drama, and it is irresponsible.

It is not based on facts, nor is it based on science. It is wrong.

Jo Ling Kent:

So you think they are wrong?

Jensen Huang:

The narrative is wrong, because it has no scientific basis.

But their concerns themselves are not wrong. I want to make this very clear: the core concerns are not wrong.

(For example) I just flew here on a plane. I am very glad that the entire aviation ecosystem takes safety so seriously. In fact, in the aviation industry, there are probably more people responsible for aviation safety, aircraft safety and flight safety than there are people responsible for designing new aircraft.

AI is just going through the same process right now.

Why not create an AI equivalent of the FDA or FAA?

Jo Ling Kent:

You have said in the past that we do not need new laws, nor new regulation. But you just said that you support various safety guarantee mechanisms in important industries. Why can't we establish an agency similar to the FDA or FAA for AI?

Jensen Huang:

Before we discuss new laws and new regulation, we should first make full use of the existing laws and regulations.

We live in a country ruled by law. The reason why we can feel safe and orderly is that we already have a large number of legal and regulatory systems.

Different application fields, whether it is food or air travel, have corresponding rules. So for AI, what exactly is missing right now?

Tell me first what is missing.

Jo Ling Kent:

I am curious about your view. Why do you think there is no regulatory gap at present? People like Dario Amodei and Sam Altman have explicitly stated that they want to see more regulation.

Jensen Huang:

My earlier response was mainly to the fact that one of the AI labs has had several cybersecurity incidents, and the other has also had some recently.

There are already many laws that can handle these incidents themselves.

There are cybersecurity laws, laws against unauthorized access to computer systems, and liability for damages. There are various liability systems for cybersecurity itself, and various liability systems for products.

If your product causes harm; if the product does not operate as you promised; if you sign service contracts with enterprise customers, there are already a large number of laws covering all these things.

My point is: Enforce these laws first.

Do not let the so-called apocalypse narrative become a reason for some companies to escape existing legal liabilities.

If you look carefully at what these companies are actually saying, read between the lines, many times they are not asking for more laws. What they may be asking for instead is to be exempted from part of the already existing legal liabilities.

I think that is a problem.

We already have many laws, let's use them first.

Do AI companies want to pass responsibility to regulators?

Jo Ling Kent:

So are you saying that these companies are asking for regulation, to some extent, because they do not want to bear all the responsibility themselves, and do not want to face what they are creating independently?

Jensen Huang:

They are transitioning from laboratories to product companies.

From a research organization with no customers and no revenue, to a commercial company that faces customers and generates revenue. And this transformation must be very drastic, because it is happening so fast.

These are some of the most influential technology companies in human history. They have extremely excellent, extremely talented researchers doing very outstanding work.

But that does not change the fact: they are moving from laboratories to genuine product and service companies.

They are already generating tens of billions of dollars in revenue, and are expected to reach hundreds of billions of dollars in the future. The speed at which they are expanding computing power capacity is unprecedented in the history of human commerce.

And they deserve it. Because the products are really good, and really useful. People want more, they of course want revenue to continue growing, and investors want to see growth. None of these things are problematic.

But product liability law still exists.

Now we should use these product liability laws to face the incidents that have already occurred, and see what happens. Do the obvious things first, do the practical things first, and then we can discuss theoretical issues. Do not skip over dealing with real problems to discuss theoretical disasters first.

"These are all young people who are first-time CEOs"

Jo Ling Kent:

So to be precise, you do not currently support adding new government regulation?

Jensen Huang:

What problem does new regulation need to solve? That is the premise. Before we do not know what exactly is missing in the existing system, why add more laws first?

Jo Ling Kent:

Then let me rephrase the question. Compared with adding new regulation, do you trust the four or five CEOs who currently control these technologies more? Do you believe they will do the right thing?

Jensen Huang:

They certainly have the ability to do the right thing. I can understand what they are going through right now.

First of all, they are all very young, and all first-time CEOs. I have been a CEO for 33 years, so I have the advantage of hindsight, sitting in the back seat here to comment on how they run their companies.

But the truth is, they are experiencing many things at the same time. Building a company, building an organization, building products, building a brand new industry, building business models, and even preparing for an IPO.

All these things are new to them, and they are all happening at the same time. I fully understand how enormous the challenges they are facing now are. At the same time, I have great confidence in the capabilities of their teams. These are all very outstanding people.

I trust these people, and I understand the environment they are in. However, I do not agree with using apocalyptic narratives to distract attention. I want them to focus on the things that I know they are actually doing.

Their engineers certainly know that the security incidents that have already occurred mean that they need to build more secure sandboxes, better isolation mechanisms and better containment systems. I believe they have also realized that during the testing and evaluation process, they need to monitor software operation more closely.

If their software is to leave their own cloud and interact with other external clouds and services, they should inform the relevant parties. If a security incident occurs, it should also be shared publicly, so that the entire industry can learn from it, and everyone can improve their own systems.

The cybersecurity industry has always worked this way.

Right now, there are cyberattacks happening all over the world all the time. At the same time, engineers all over the world are frantically patching software, improving software, and fixing vulnerabilities.

These are all extremely complex systems. We have open software, open source software, and enterprises can defend against each other and update their own systems. This is one of the important reasons why the entire industry can operate.

Everyone looks out for each other. Defenders have the advantage in numbers, with hundreds of companies forming a community to face attackers together.

So there are of course many things worth analyzing in these security incidents. But the core fact has not changed: these companies are now returning to engineering itself. They will improve their engineering, and invest more in evaluation, benchmarks and testing. This is a good transformation.

Nvidia makes money from AI, why should the public trust you?

Jo Ling Kent:

When Americans see you, they see the founder and CEO of Nvidia, and also see one of the most valuable companies in the world.

On the specific issue of AI safety, why should the American public trust your judgment? Your company's success and your personal wealth are directly related to selling AI chips.

Jensen Huang:

Our company's success is directly tied to one thing: these products and services can be deployed safely.

If the computer industry cannot create value for the world safely and efficiently; if we cannot continue to generate huge positive impacts, then our value will naturally decline.

So whether from the perspective of personal interests or the interests of the entire industry, we have extremely strong incentives to ensure that this new category of applications, namely AI, can be produced safely and delivered to the world.

We have to do it. We must maintain a responsible optimism. We do this job because we believe it is beneficial to the world. We want our children to live in this future, and my children are already in this industry.

We want a future with less human suffering, more efficient enterprises, and faster knowledge discovery.

We want to solve the problem of population aging, have sustainable energy, and even use the market power generated by AI to push the world into the era of sustainable energy. We all want these futures. But we must do it responsibly, we must do it safely, and we must do it maturely.

We cannot scare the public with dramatic, apocalyptic future predictions. Because that is not true at all.

China is the elephant in the room

Jo Ling Kent:

There is an unavoidable issue here, which is competition with China. A very common view is that the United States cannot slow down AI development, otherwise it will lose to China. Do you think this logic holds?

Jensen Huang:

If I could give advice to every CEO, every industry, every student, every professor, and everyone in American society, I would want the United States to be as prosperous as possible.

The more prosperous and wealthy our economy is, the more we can support social security, the more we can support national and military security, and the more capable we are to be a responsible global citizen, helping other countries face challenges, difficulties and suffering. The more capable our country will be to carry out charity and aid.

So I would advise everyone in America: do your utmost in the same direction.

Jo Ling Kent:

So should we slow down? Will slowing down affect the United States' competition with China?

Jensen Huang:

The United States should advance as fast as possible, regardless of what other countries do.

If China discovers a cure for cancer, that is good for the United States too. If China invents a new sustainable energy technology, that is good for the United States too. If China leads in certain fields of science and mathematics, we can learn from them and benefit from it, that is also a good thing.

But remember, the United States itself has extremely powerful capabilities. This is a country with extraordinary innovation capabilities. I am very proud to be part of the American technology industry.

I think the United