Latest speech by Jack Clark, co-founder of Anthropic: The Singularity has arrived, and AI will change the development trend of the global economy.

In a recent public speech, Jack Clark, co-founder of Anthropic, put forward a staggering judgment: The Singularity has arrived. He believes that this year we have obtained convincing evidence that AI systems can not only accelerate scientific progress, but even begin to surpass human experts in certain fields. The core viewpoint can be summarized as: AI is reshaping national security and economic trajectories, and driving its own development; science is the biggest beneficiary of AI; but we must plan for success and strike a balance between acceleration and restraint.

I. One person spent 40 years preparing for this speech

I am that person. It took me roughly 40 years to stand here and deliver this speech.

My entire career has been defined by one thing: finding information, then sharing that information with the whole world. Twenty years ago, I started my career as a journalist covering the data center sector. It was through that work that I developed a strong interest in artificial intelligence. In 2016, I left the journalism industry and joined an open-source AI company. A few years later, I co-founded Anthropic.

Since then, I have been trying to tell the whole world that one day, AI systems will become so powerful that they may transform national security, change the economic trajectory of the planet, and perhaps eventually gain the ability to develop on their own.

Well, that moment has arrived.

AI is transforming national security, reshaping economic trajectories, and driving its own development. Many of you are aware of this. You have to treat AI as one of the many destabilizing forces that are shaping your world and your work. I am here to explain that: AI is moving toward bringing far greater changes beyond our current expectations, and we should prepare accordingly.

II. The Singularity has arrived: Progress itself will create more intelligent entities

For many years, researchers and science fiction writers have predicted that AI systems will lead the future to glory, greatly accelerate the overall pace of scientific and technological progress, and things will develop so fast that humans may struggle to keep up. They call this the Singularity.

I believe that this year, we have obtained convincing evidence that all of this has already begun.

Writer Vernor Vinge predicted the Singularity in 1993. He said that when forces exceeding human intelligence drive progress, this process will become even faster. In fact, there seems to be no reason why progress itself will not create more intelligent entities within a shorter period of time. He described the Singularity as a critical point for our models: We must abandon old concepts and embrace the new reality.

The Singularity has arrived. Its impact will spread to every aspect of life. But its most important implication is very simple: We should expect that in the next few years, not just AI, but the development speed of the entire scientific field will accelerate. This will be confusing and disorienting. It will bring enormous wealth, but it may also bring huge dangers.

Today I will focus on the first point, but I will not ignore the second. I will talk about its benefits, the rapid progress, and what we need to do to share these benefits.

III. AI-driven scientific discoveries will change the direction of the global economy

Many of the benefits that AI will bring to humanity stem from science. I believe that AI-driven scientific inventions will change the direction of the global economy, transform the way countries plan and build infrastructure, make us healthier, allow us to spend more time with our loved ones, and at the same time prompt us to think about profound philosophical questions about choices.

Over the past half century, most countries have faced the problem of declining productivity of scientific researchers. At the same time, the number of researchers has grown rapidly. Creativity and inventions seem to be increasingly difficult to come by. But something important has changed: We have now built systems that can enhance, expand, and partially automate human labor.

This is most evident in the field of software development. AI systems are now more capable and far more efficient than the vast majority of software engineers in the world. The software development sector is changing: now one single person can write software for anything they can imagine, creating things that previously required an entire team months to complete.

Inside Anthropic, we have already seen concrete evidence. Our researchers now write almost 10 times the amount of code they did in the first four years after the company was founded. And this growth in coding is basically a direct reflection of the improvement of AI capabilities.

But new AI features will also bring new challenges. The improvement of coding capabilities means that AI systems have also made significant progress in cybersecurity. While the world is racing to deploy these powerful capabilities to build cyber defense systems, criminals are also racing to use these systems to carry out hacking attacks. These things are inextricably linked. Advances in AI in fields such as programming will have both positive and negative impacts in fields of international significance such as cybersecurity. This indicates that future progress will bring new capabilities and new challenges.

And these capabilities are not limited to programming. The same thing will happen in the fields of science and robotics. This is already starting to happen. This means that some important characteristics of the global economy may change, such as the speed of innovation, or the frequency at which we invent entirely new general-purpose technologies. The Internet, mobile phones and AI are all general-purpose technologies that have created new industries around themselves.

IV. Evidence of scientific acceleration: From mathematics, biology to robotics laboratories

In 2026, we have already seen initial signs of AI: these systems may be able to conduct cutting-edge scientific research.

In the field of mathematics, AI systems have been used to propose solutions. Long-standing mathematical problems such as the Navier-Stokes equations Millennium Prize problem, the Jacobian conjecture, and several unsolved Erdős problems, in many cases, AI systems have provided core insights that made solutions possible.

In biology, researchers use AI to design viruses from scratch and identify new candidate drugs for testing. There are currently drugs undergoing clinical trials, and these products are largely designed by AI systems.

In the field of robotics, AI systems are now increasingly capable of operating laboratory equipment autonomously. This allows us to further automate experiments, from biology to materials science, and make faster progress in fields such as batteries or fuel cells.

Viewed in isolation, these examples are all exciting. But the fundamental reason is that they are all related to the same underlying technology. For most of these use cases, the same single underlying cutting-edge model is adopted. What we have now is a single entity, AI systems like Claude, ChatGPT or Gemini, that are capable of conducting research work at the forefront of multiple scientific disciplines. We have virtual polymaths.

There is more quantitative evidence. One indicator is the score of AI systems in a test called "The Last Exam for Humans". This ominously named test covers a wide range of topics including chemistry, computer science, mathematics, classical literature and ecology. When it was launched in January 2025, the highest-scoring AI system only got a score of 8%. As of this week, the highest-scoring system, Claude Opus 5.5, scored 67.7%.

Another example is Frontier Math, released in November 2024, designed to assess the ability of AI systems to solve hundreds of challenging mathematical problems carefully curated and evaluated by professional mathematicians. When it was first launched, the best systems only got a score of 2%. Now they can get 94%.

More broadly, if you pick different fields of science, you will find the same pattern: Progress is accelerating, especially over the past two years.

V. One signal: AI not only answers questions correctly, but also corrects human experts

The latest example comes from a paper published last week. This paper tests the ability of AI systems to solve unpublished theoretical physics problems. The 50 problems in the paper were written by physicists from all over the world and have not been published, with answers derived from their own research.

Unsurprisingly, the AI systems got high scores. But what was surprising was how they got those high scores.

I would like to quote Suat Raju, an Indian physicist and one of the authors of this test. He commented on the results on Facebook. He said that initial assessments suggested that GPT6 Astra had the highest score in the late 80s, closely followed by Opus 5.5. But when we began to investigate the possible mistakes the models might have made, the models, especially Astra, insisted that they were right and the answers provided by the experts were wrong.

When we consulted the experts again, it turned out that the experts had found subtle errors deeply buried in their calculations in multiple cases. When they revised their answers, Astra's score began to rise slowly, finally settling at 98%, with 49 out of 50 answers correct.

Think about this carefully. Some of the top scientific experts in the world designed a test with questions that AI could never have read. When they tested the AI, it not only scored highly, but also achieved an even higher score by finding errors in the human-provided answers.

There is usually a scene in movies, usually two-thirds of the way through the film, where the young protagonist finally proves to their mentor that they have the qualities to go out into the world and achieve success on their own. This is a common narrative device: the teacher becomes the student, or the student becomes the teacher. This is exactly what happened last week.

VI. Why this matters: Creativity is the final frontier, and scientific breakthroughs will trigger explosive growth

Why am I telling you all this? Why is the continuous progress of AI systems in the scientific field important?

Because doing real cutting-edge science that can make breakthroughs and expand the boundaries of knowledge is actually testing the final frontier of AI capabilities: creativity. My point is that if AI systems can solve problems creatively and succeed in the scientific field, we should take this as an indication that they have the ability to extend research results to multiple fields, including of course AI research itself. This could have an explosive impact on innovation.

Therefore, if AI systems become better at science, this will greatly drive the growth and landscape of the future economy. This means that you can make decisions today to prepare your country and people to benefit from this trend in the future.

One of the main sources of growth is research, which generates new ideas. One of the drivers of growth is our stock of human capital: people all over the world who ask questions, conduct scientific research, and help move the world forward. But we also know that, as some economists say, good ideas are getting harder to find. The cost of breakthroughs is rising, and we have to invest more and more research efforts just to maintain the same pace.

So my question is: What if the situation has changed? What if this technology allows us to change the number of researchers on the planet? How long will it take to train them?

As I said when I listed my resume at the beginning, it took me roughly 40 years to get to where I am today. What would it mean if hundreds or thousands of copies of me appeared at the same time? What if a new version of me could be created immediately without waiting 40 years, how would my work change? What if I could replicate myself and order them to work under my supervision?

In the next few years, most sectors of the economy will face this situation. We will be able to enhance human labor and conduct research in disciplines such as frontier science in ways we have never been able to do before. This will likely change the speed at which we discover good ideas, because we will be able to increase the number of researchers by hundreds or thousands of times. When encountering any problem, by empowering our human scientists, we can also get the work done quickly.

If you want to cultivate a new generation of researchers in a certain field, it may take 10 to 15 years of targeted education and millions of dollars for a small group of people. On the other hand, these AI systems can be created immediately. As long as there is enough computing power, the number can be arbitrary. They can be used across fields, allowing people to benefit from new scientific insights almost instantly. They can combine existing knowledge in novel and unexpected ways to make discoveries.

This means we can expand rapidly: Everyone on Earth can work side by side with virtual scientists. Therefore, in the next few years, every scientist on this planet will become increasingly important and powerful, as they gain the ability to manage hundreds or thousands of like-minded experts to study any problem that interests them.

VII. 44 trillion US dollars: Economic growth may accelerate

The team of economists at Anthropic has modeled this. We tried to simulate different versions of the economy based on different trajectories of AI technology.

If you assume, as people in laboratories do, that AI models will be able to operate autonomously, complete a wide variety of tasks, and spread over the next few years, economic growth will accelerate. Then, under these Singularity conditions, one model predicts that a huge economic transformation will occur: US GDP may reach 44 trillion US dollars by 2030.

I know this sounds outrageous. But it is closely tied to the continuous advancement of technology. If GDP grows to this level, then traditional economic theories will no longer hold. This suggests that emerging industries and inventions need to establish some mechanism to support such a large-scale economic expansion. And that is exactly the implication: through this concept, AI systems are capable of conducting independent scientific inquiry.

So what should we do? How can we make the whole world benefit from these evolving capabilities? How can we balance this progress opportunity with appropriate prudence?

VIII. Prepare for success: Infrastructure, supply chains, and scientific facilities

We all need to plan for success, so that humanity can use this technology to create a life of abundance. We need to be prepared to make informed choices about how to use this technology itself. Just because AI will become an excellent scientist in the future does not mean that we should absolutely use it for everything.

When it comes to how to plan for success, I will start with my own personal experience. Over the past decade, I have been dealing with governments around the world. I have tried to talk to people all over the world about this. I told them that as AI systems become more powerful and more useful, one of the best things they can do is position themselves to take advantage of this upcoming revolution: Build sufficient power capacity and data center space to be able to host powerful AI systems of the kind discussed here, and invest in the AI supply chain.

Few policymakers have taken this to heart. I think now we have to face the consequences. In today's world, data center capacity is concentrated in the United States, China, and increasingly the Middle East, and the supply chain is highly concentrated. Today, this kind of construction is still very important. But this is only the first phase of the economic and scientific revolution. The continuous advancement of AI suggests that there is still more to build now.

I believe that more countries can get involved if they wish. Because the next phase of the AI revolution is the expansion into the physical world. Specifically, by allowing AI systems to conduct scientific experiments and sometimes operate robots together. Therefore, just as we did with computing and electricity before, we will want to deploy more scientific methods, equipment and robotic systems. Otherwise, we will soon hit capacity limits, and history will repeat itself.

So we need to upgrade our capabilities now to seize the upcoming opportunities. According to UNESCO estimates, there are roughly 10 million scientists in the world. Even with only 10 million scientists, we are far too few to supervise the work today. There are too few laboratories capable of conducting materials science analysis, too few biological laboratories to develop and test new drugs, and too few large scientific instruments.

Tell me, what would happen if instead of having 10 million scientists, we had 1 billion virtual scientists guided by our own 10 million users? This is what the world will unsurprisingly look like as it evolves toward us. We must be prepared.

If we are prepared, we will be better positioned to translate intelligence and AI systems into benefits for all of us: longer and healthier lives, more energy, more food, and more wealth. Just like data centers, the enterprises in various countries that build and develop this supply chain will benefit the most. That is a choice, and an opportunity for everyone.

If you are a developed country with a rich scientific ecosystem, you have the opportunity to make all your scientists far more productive. If you are