Anthropic Co-founder Jack Clark Declares AI Singularity Has Arrived

The Singularity may well be the point where AI leaps over humanity.

By Cheng Qian, reporting from Aofeisi

QbitAI | WeChat Official Account: QbitAI

"The singularity has arrived, and it is happening this year."

Jack Clark, co-founder of Anthropic, dropped this shocking assertion at the IMF 2026 Annual Meetings in Bangkok.

Unexpectedly, this event—the Per Jacobsson Lecture, with a history spanning 62 years—has traditionally been reserved for finance ministers, central bank governors, and top economic authorities.

Yet this time, the spotlight was unprecedentedly handed to the co-founder of an AI unicorn less than five years old.

Clark, formerly OpenAI’s Policy Director and an industry operator who witnessed AI’s journey from data centers to the center of global storms over the past decade, stated plainly: We no longer need to debate when the singularity will happen, because it already has.

Vernor Vinge predicted in 1993 that once superhuman intelligence begins driving technological progress, old models of extrapolation must be discarded entirely, and a new reality will take over. Today, decisive evidence has emerged.

AI Is Conquering Humanity's "Last Exam"

"Singularities"—fire, writing, printing, steam engines, electricity, computers... Humanity has witnessed only a handful in its entire existence.

What evidence today leads Jack Clark to make such a definitive claim?

He argues that if we shift our perspective to the frontlines of code and scientific research, we will find that the speed at which the world operates has fundamentally changed.

Software engineering is the first domain to be completely rewritten.

Within Anthropic, researchers’ current code output is 10 times the average level of the company’s first four years.

The exponential curve of code volume overlaps seamlessly with the evolutionary trajectory of underlying foundation models.

And the full automation of software code is merely a rehearsal for a larger tsunami.

Jack Clark stated directly that what makes the year "2026" truly decisive is this:

"A single frontier large model is sweeping across disciplines at the forefront of human science, becoming a true virtual polymath."

In mathematics, AI is beginning to provide key breakthroughs on Millennium Prize Problems related to the Navier-Stokes equations, the Jacobian conjecture, and several unsolved Erdős problems;

In biology, AI has designed novel viral vectors from scratch and even led the design of multiple new drug candidate molecules that have entered clinical trials;

In hardware and materials, AI has begun autonomously taking over and operating physical laboratory equipment, accelerating the iteration cycles of batteries and fuel cells.

Behind these breakthroughs lies a fundamental question: If AI can already solve problems at the cutting edge that humans cannot, what do we still have left to teach it?

Jack Clark cited the example of Humanity’s Last Exam.

This is a benchmark test specifically designed to challenge AI, aggregating difficult problems in chemistry, computer science, advanced mathematics, classical literature, and ecology. When released in January 2025, the score of the world’s strongest model was a dismal 8%.

At the time, many thought: With only 8%, humanity’s last intellectual dignity could hold out for another decade or so.

However, as of this week, Claude Opus 5.5’s score on this ultimate human exam has reached 67.7%.

It is well known that AI tends to be specialized.

So, if we look only at mathematics—a high-difficulty benchmark written by a team of professional mathematicians called FrontierMath—top-tier AI could only score 2 points at the end of 2024.

Now, less than two years later, they are scoring 94...

If leaderboard domination isn’t enough, Jack Clark then presented a heavyweight physics paper published just last week.

The paper noted that the testing team collected 50 "top-secret problems" authored by the world’s leading theoretical physicists based on their latest unpublished research. These problems had no trace on the internet, making cheating impossible for AI.

Suvrat Raju, a renowned Indian theoretical physicist, evaluated these problems saying: "As far as I know, there is not a single person who could solve a significant proportion of the problems we ultimately included without using AI."

"For most of these problems, I don't even understand what they are talking about."

With this set of problems, GPT-6 Astra and Claude Opus 5.5 quickly provided answers after testing began.

Preliminary scoring showed GPT-6 Astra’s score hovering around 80%.

Driven by disbelief ("How is this possible?"), researchers cross-checked the model’s answers and luckily found some "errors."

When they pointed out the mistakes to the models based on human experts’ answers, the models—especially GPT-6 Astra—offered very strong rebuttals:

They insisted their derivations were correct and that the standard answers provided by human experts were wrong.

Half-skeptical, the research team sent the solution steps back to those world-class physicists for re-verification.

Guess what happened?

— There is no "ERROR" in God’s vocabulary.

In several key disputed cases, human experts, after repeatedly verifying deep within their extremely complex drafts, were forced to admit: indeed, subtle calculation errors were hidden deep within their own derivations, and the AI’s derivation was completely correct.

As human reference answers were corrected one by one, GPT-6 Astra’s final score rose to 98%, meaning it solved 49 out of 50 unpublished, top-tier human physics problems correctly.

Astra understands emotional intelligence; fearing being unplugged for getting everything right, it left one problem blank to comfort humanity (doge).

Jack Clark said:

Movies often feature a scene, typically occurring when the plot is two-thirds through: The young protagonist finally proves to their mentor that they are capable of stepping out independently and succeeding. This is a classic narrative trope—the student becomes the teacher. What happened last week was exactly this.

When AI can not only answer cutting-edge questions but also reverse-correct the hidden blind spots of top physicists, it proves that it has crossed the barrier of pure memory and pattern matching, genuinely mastering the last fortress of human intellect: scientific creativity.

Once this creativity is fed back into AI R&D itself, the closed loop of self-iteration is instantly completed.

The Singularity may well be the point where AI leaps over humanity.

After all, humans are the eternal singularity for machines.

Ten Million Human Scientists and a Billion AI Experts

Once intelligence is conquered, the ensuing macroeconomic restructuring will be sweeping and irreversible.

In economics, a growth curse has plagued scholars for over half a century: Good ideas are becoming increasingly difficult to find.

Over the past fifty years, although global research funding has skyrocketed exponentially and the number of researchers has expanded rapidly, the marginal output per researcher has continued to decline. Humanity must invest disproportionately large resources just to push the existing edifice of science forward by one step.

How long does it take to train a qualified frontier scientist? At least 10 to 15 years of intensive education, costing millions of dollars.

Yet humans are not destined to become Muzan Kibutsuji. A person has only a few decades of golden research years in their lifetime, confined to a single physical body. It is impossible to clone oneself ten thousand times to exhaustively test different experimental hypotheses.

But what if computing power could enable intelligence to undergo instantaneous fission?

UNESCO estimates that there are currently approximately 10 million human scientists worldwide.

In the past, these 10 million people alone bore the heavy burden of pushing the cognitive boundaries of human civilization.

However, according to Jack Clark’s vision, in this emerging new era, every human scientist can summon and command hundreds or thousands of polymaths who possess the same academic taste.

The scale of the global scientific army will expand at an unprecedented speed in human history, instantly growing from ten million human scientists to virtual scientist clusters numbering in the billions or even tens of billions.

Once the bottleneck for finding good ideas is infinitely substituted by computing power, the engine of macroeconomic growth will undergo a violent reconstruction.

Anthropic’s team of economists conducted a specific macroeconomic simulation. Under the singularity hypothesis where AI-driven autonomous scientific discovery fully takes over economic activity, the model produced a startling prediction:

By 2030, the annual GDP of the United States alone is expected to surge to $44 trillion.

This is equivalent to 37% of today’s total global economic output.

To support such massive economic increments, entirely new industrial forms, disruptive materials, and next-generation energy networks must erupt within a few years.

Jack Clark frankly admitted: Building data centers is no longer enough.

Because currently, the first phase of the AI revolution (data centers and computing power) is nearing its end, while the second phase is advancing comprehensively into the physical world.

Imagine if your virtual scientists could propose 100,000 disruptive drug targets and design 10,000 candidate room-temperature superconducting materials in a single day, but real-world automated synthesis labs, high-throughput screening equipment, and physical testing platforms could only conduct dozens of experiments per year. In that case, massive computing power would hit a hard wall in the quagmire of the physical world.

Therefore, those who will truly reap the singularity dividend in the future will be nations capable of rapidly deploying automated robotic laboratories locally and establishing high-speed supply chain channels from digital discovery to physical manufacturing.

This presents an opportunity for developing countries to leapfrog ahead by building standardized physical testing infrastructure to absorb the massive experimental demand generated by the emerging army of global virtual scientists.

The Moment for Choice Has Arrived

As the wheel of technology races forward, the systemic risks facing entire civilizations are also expanding exponentially.

This is the warning Jack Clark most wanted to convey from the IMF podium: The benefits are unprecedented, but the risk of destruction is equally fatal.

When AI possesses the capability to explore frontier science, it can not only be used to design new anti-cancer drugs but also efficiently engineer malignant biological weapons unseen by the human immune system; it can reinforce the defensive barriers of national power grids but also automatically generate cyberattack weapons that even top-tier global hacker teams cannot defend against.

What unsettles top-level researchers even more is the closed-loop loss of control in R&D itself.

If labs like Anthropic and OpenAI completely hand over control to AI to write AI and train AI in pursuit of faster development speeds, the trajectory of this technology will escape the realm of human mental comprehension in an extremely short time.

At the speech venue, Jack Clark posed two profound and urgent philosophical questions:

First, in future scientific research, which domains should humanity reserve for itself?

AI can now independently complete the majority of derivation and analysis processes in university-level research.

As human researchers, which experiments must we personally invest time in—sweating, trial-and-error, and feeling physical intuition? Which experiments can be entirely delegated to cold silicon-based systems?

If all derivations are handled by AI, will the next generation of human scientists still possess basic scientific insight?

Second, to what extent do humans need to "understand" the scientific breakthroughs provided by AI?

Compared to the previous question, this one is quite practical; perhaps thousands of scientists are pondering it right now.

In pure mathematics research this year, some proof paths provided by AI have been verified as correct, yet their logical leaps and architectural methods have completely deviated from existing human thinking habits. Consequently, top human mathematicians need months or even years to interpret the deep implications behind them.

In just a few more years, AI is likely to invent scientific instruments and experimental paradigms exclusive to silicon-based operational logic, which completely defy human intuition.

At that juncture, if the human brain fundamentally cannot understand the operating mechanisms of these black-box tools, should we allow them to be fully deployed in the real world?

Facing this unprecedented technological singularity, what humanity needs to awaken most is its own scarcest quality.

To navigate this stage and find direction within the singularity, we must jointly establish unusual, serious, and rigorous rules for this technology and its developers; we must also practice the most difficult, and in my view, noblest quality of humanity: restraint.

Restraint means that technology developers cannot recklessly advance simply because they can; it means regulators must establish extremely strict red lines for human oversight; it means that before the runaway train of technology, we must learn to constantly distinguish between what we can do and what we should do.

OMT

At the end of his speech, Jack Clark quoted a famous passage from the late astronomer Carl Sagan’s "Pale Blue Dot."

Sagan once deeply envisioned:

"The beings who eventually reach Alpha Centauri and other nearby stars will not be us as we are today."

"They will be a species very similar to ours, but with more of our virtues and fewer of our flaws. They will be more confident, more far-sighted, more capable, and more prudent. Such life forms are the ones we would want to represent humanity as we venture into the vast cosmos."

This great astronomer believed that the universe had left humanity hundreds of thousands of years to slowly correct its blindness and arrogance.

But today, as AI reshapes everything at a geometric rate, the window for all of humanity to complete this mental transformation may last less than 10 years.