In short:
- Altman's statement on July 25, 2026, is not an isolated event but part of a broader wave: that same summer, similar statements were made by Musk, Huang, and earlier by Hassabis.
- Independent researchers (Stuart Russell, Roman Yampolskiy, Nick Bostrom) almost unanimously deny that the singularity has already arrived, although they acknowledge the acceleration of progress.
- Researchers propose a clear test: true singularity should leave "footprints" in four places – independent AI scientific discoveries, macroeconomic productivity, dissemination of benefits beyond specific industries, and a continuous cycle of "machines creating stronger machines."
- None of these four tests have been convincingly passed as of mid-2026.
Contents
Altman's statement as part of a broader wave, not an isolated event
When Sam Altman said on the Relentless podcast on July 25, 2026, that humanity "is already in the singularity," the media picked it up as a separate sensation. But looking at the summer of 2026 as a whole, it becomes clear: this is not an isolated statement from one CEO, but part of a broader chorus. During the same period, similar statements were made by Elon Musk and Jensen Huang, and earlier, in May, by Google DeepMind CEO Demis Hassabis. Writer James Barratt also publicly stated that the unpredictable, civilizationally significant development of AI already meets one of the influential definitions of singularity (Forbes).
This article is not a retelling of the news about Altman (more details about the incident itself and the chronology of that day are in our news: Sam Altman stated that humanity has entered the era of singularity), but an attempt to understand why several industry leaders started talking about it almost synchronously in the summer of 2026 – and what independent researchers, not just CEOs of commercial companies, actually think about it.
Demis Hassabis's position: "foothills of the singularity" (May 2026)
Chronologically, Hassabis was the first of the big three to talk about this – two months before Altman's July statement. But to understand why his words are worth taking seriously, it's important to know the context: unlike many other voices in this discussion, Hassabis is not just a CEO, but a practicing researcher with a PhD in Cognitive Neuroscience, a co-founder of DeepMind in 2010, and the person whose team is behind AlphaFold – a system that solved a 50-year-old problem of protein structure prediction and earned him the Nobel Prize in Chemistry in 2024. This means that when Hassabis talks about the pace of AI progress, his words are backed not only by a CEO's marketing position but also by a real track record of scientific breakthroughs.
That's why it's telling how much more cautiously he phrases his position compared to Altman. At Google I/O in May 2026, Hassabis called the current moment "foothills of the singularity" – a deliberately figurative and indirect metaphor: the foothills are not the summit yet, but only the point from which the summit first becomes visible. This is a fundamentally different statement than Altman's categorical "we are already in the singularity": Hassabis says "we see where we are going," not "we have already arrived."
Later, speaking at the Indian AI Summit, Hassabis further developed this idea, comparing the potential of AGI to fundamental civilizational shifts like the discovery of fire or the harnessing of electricity, and predicting an impact ten times greater than the Industrial Revolution within the next decade. But even here, he added a key caveat – "provided that the development of technology is managed responsibly" (Finviz/AFP). This caveat is not a rhetorical flourish: it directly implies that the scale of the technology's impact does not guarantee a positive outcome on its own, and it is the management of the process that will determine whether the optimistic scenario comes true.
What Elon Musk and Jensen Huang said
If Hassabis's position is trustworthy due to his scientific track record, then Musk's and Huang's statements should be approached with the opposite, more skeptical filter – and here's why.
Elon Musk essentially echoed Altman's phrasing almost verbatim, writing that same week that humanity "has entered the singularity, albeit in its early stages" (Gary Marcus). The problem is that Musk has a long and well-documented history of public predictions about AI and autonomous technologies that have regularly failed to materialize within the stated timelines – the most famous example being the annual promises of "fully autonomous" Tesla cars, which have been postponed year after year since 2016. This doesn't mean that the current statement is necessarily false – but it's a significant reason why researchers tend to demand separate, independent confirmation of his predictions rather than accepting them at face value.
NVIDIA CEO Jensen Huang, during the same period, announced the achievement of AGI – although, unlike Altman and Musk, he consciously avoided the word "singularity," focusing on a narrower technical assertion (Gary Marcus). Here, it's worth noting an obvious, though rarely voiced, nuance: NVIDIA is a company whose business directly depends on how much the industry and investors believe in the proximity of a qualitative leap in AI capabilities, as it is this belief that justifies colossal capital expenditures on purchasing NVIDIA chips. This doesn't make Huang's statement automatically false, but it is a structural conflict of interest that should be kept in mind when evaluating the weight of his words – just like the words of any CEO who directly benefits from increased demand for their own products.
Together, these two statements offer an interesting contrast to Hassabis's position in the previous section: where a scientific track record supports cautious phrasing, here commercial interest and a history of unfulfilled predictions demand greater caution from the reader themselves.
Dario Amodei's (Anthropic) position
The position of Anthropic's CEO stands out from other CEOs not because it's softer, but because of how long and consistently it has been articulated. While most of his colleagues only started talking about the singularity in the summer of 2026, Amodei has been warning about the approach of powerful AI for several years – as early as his extensive essay in October 2024, "Machines of Loving Grace," he predicted the emergence of what he called a "genius country in a data center," around 2026-2027: a system capable of independently performing the work of an entire institute of Nobel laureates simultaneously, and compressing 50-100 years of progress in biology into 5-10 years (analysis of Amodei's essay). That is, unlike Altman, for whom the July statement sounded like an unexpected shift in rhetoric, for Amodei it is rather a logical continuation of the position he has publicly advocated for two years.
Another significant difference is the tone of the message itself. Amodei stubbornly avoids statements like "we are already in the singularity" – instead, he always frames it as a warning, not a triumph. A telling quote from his recent public speech: "Humanity is about to be handed almost unimaginable power, and it is deeply unclear whether our social, political, and technological systems possess enough maturity to wield it" (TechRadar). The difference in phrasing here is not cosmetic: Altman essentially states a fact with a hint of excitement, while Amodei formulates a warning with a hint of anxiety – both CEOs agree on the speed of progress, but diametrically disagree on how it should be discussed publicly.
It is telling that in the spring of 2026, Amodei went beyond general warnings and publicly acknowledged a specific technical fact: Anthropic's models can already independently build tools and "scaffolding" to improve their own workflow – and this is essentially an early, limited form of the recursive self-improvement he warns about in his essays (36Kr). This is not an abstract acknowledgment: it directly echoes Anthropic's April research case, where Claude-based agents independently completed 97% of an AI safety research task – meaning Amodei is not talking about a hypothetical future, but about what he is already observing in his own labs. For more on this experiment and why the company itself still calls the full RSI cycle "not yet inevitable," read our technical analysis: Recursive Self-Improvement.
What Stuart Russell Says
If the CEOs of commercial labs are a party interested in making their technologies look as groundbreaking as possible, then Stuart Russell is one of the few voices in this discussion who has nothing directly to sell. He is a professor at the University of California, Berkeley, and co-author of the classic textbook "Artificial Intelligence: A Modern Approach," which the vast majority of computer science students worldwide still study; it is this academic status that makes his commentary such a weighty counter-voice to the CEOs' statements.
Russell gave perhaps the most concise and sharpest answer to the direct question of whether the singularity has arrived: "No, and Altman doesn't think so either" (Forbes). This answer is intentionally constructed as a logical trap, not just a denial. Russell's argument is specific and verifiable: Altman himself publicly predicted that AI systems would only be able to perform "a significant fraction" (not all) of OpenAI's research by March 2028 – that is, more than a year and a half after the July statement about "the singularity has already arrived." Russell points out this direct contradiction: if the necessary level of capabilities, according to Altman's own assessment, will only emerge in years, then the current public statement logically cannot describe the present moment (AOL/Business Insider).
Nick Bostrom's Perspective
Nick Bostrom's position is worth reading separately from the other voices because he is the one who set the vocabulary that this entire discussion uses today: his 2014 book "Superintelligence: Paths, Dangers, Strategies" was the first to systematically lay out the possible paths to superintelligence and the associated risks, and it largely determined how the industry began to discuss these issues at all – including how we define ASI in this cluster of articles (more details in our analysis of AGI vs ASI). That is, Bostrom is not an outside commentator, but one of the architects of the very conceptual framework in which Altman, Russell, and everyone else are now arguing.
This is why it is telling how moderate, rather than categorically dismissive, a position he takes, unlike Russell's sharp "no." Bostrom acknowledges the "first stirrings" that machines are already making a real contribution to their own AI research – that is, he does not dismiss the empirical observations that optimists rely on. But he points to one specific missing link, which, in his opinion, does not allow us to speak of a fully launched cycle: continual learning – the ability of a system to organically accumulate new experience and change from it in the same way the human brain does, rather than just within separate, predefined training cycles (Forbes). This is a technically more precise and narrower denial than Russell's general skepticism: Bostrom essentially says "the process has begun, but one key architectural detail is still missing" – rather than "the process hasn't begun at all yet."
Roman Yampolskiy's Opinion
Roman Yampolskiy is an AI safety researcher, an associate professor at the University of Louisville, who has specialized narrowly and specifically in the "AI Control Problem" since the 2010s – the question of whether it is even possible to guarantee control over a superintelligent system under human supervision. This sets him apart from other voices in this discussion: he is not a CEO evaluating the progress of his own product, nor a general philosopher, but a specialist whose entire career has been built on finding the answer to the question "when and how does humanity lose control" – that is, the question at the very heart of the singularity debate.
This is why his formulation is worthy of special attention: Yampolskiy has formulated perhaps the most precise and useful criterion for testing the very concept of singularity: "Rapid progress itself is not yet singularity" (Forbes). At first glance, this sounds like a simple denial, similar to Russell's position – but in essence, it is a fundamentally different, narrower, and methodologically more precise argument.
The difference lies in what is being questioned. Russell denies a specific fact: do models already have the necessary level of capabilities. Yampolskiy, however, points to a logical fallacy in the very formulation of the question: even if progress is indeed accelerating, and benchmarks are indeed breaking records every month – this is proof of speed, not proof of loss of control. Singularity, by definition, as we use it in this cluster of articles, is not just "very rapid progress," but specifically the moment when the pace of change ceases to be subject to human direction and oversight. One can imagine a world where AI improves weekly, breaks all previous records – and yet every step of this improvement is still decided, funded, and verified by humans. This would be impressive progress, but not a singularity in the strict sense.
This distinction is not abstract semantics, but a practical filter that should be used to check almost every loud statement in this discussion, including the statements of CEOs discussed above: before agreeing that a particular breakthrough "brings the singularity closer," one should ask not "how impressive is this," but "who, after this breakthrough, continues to determine the direction of development – humans or the system itself."
Where Independent Experts Agree and Disagree with CEOs
Reading all seven positions above in sequence, it's easy to get lost in the details and perceive this discussion as a chaotic dispute where everyone says something different. But if you bring them together, a surprisingly clear pattern emerges – and it is this pattern, rather than individual quotes, that best explains what is actually happening in the industry.
Virtually all participants in the discussion, including the harshest critics, agree on one fact: progress is indeed accelerating, and the autonomy of systems is indeed increasing. None of those discussed above – neither Russell, nor Bostrom, nor Yampolskiy – deny the speed of technical changes itself. They disagree not on this, but on the following, much narrower question: does this acceleration mean a loss of human control over the direction of development, or is it simply faster, but still fully human-controlled progress. This is a critical distinction, because it is this, not the pace itself, that is the definition of singularity we use in this cluster of articles.
| Who |
Has the singularity arrived |
Key argument |
| Sam Altman (OpenAI) |
Yes |
The pace of change already exceeds human predictive ability |
| Elon Musk |
Yes, in its early stages |
Supports Altman's formulation |
| Demis Hassabis (DeepMind) |
Approaching ("foothills") |
More cautious formulation, emphasis on responsible management |
| Dario Amodei (Anthropic) |
Does not state directly |
Focus on warning about power without readiness to wield it |
| Stuart Russell |
No |
Altman's own predictions contradict the "already arrived" thesis |
| Nick Bostrom |
Partial signs, not fully |
Lacks continual learning |
| Roman Yampolskiy |
No |
Rapid progress ≠ loss of control |
If you look at the table more closely, another pattern emerges besides the "CEOs vs. academics" division: the positions within each camp are also not uniform. Among the CEOs themselves, Amodei consistently stays away from the categorical "yes" of Altman and Musk – and, as we discussed above, does so consciously, choosing the language of caution over the language of triumph. Among academics, Bostrom also does not merge entirely with the firm "no" of Russell and Yampolskiy – he acknowledges partial signs of the process, albeit without a key missing link. That is, the real picture is not two monolithic camps, but rather a spectrum, where the degree of caution in formulation correlates quite accurately with how much the speaker's personal or commercial reputation depends on the fulfillment of the prediction.
Why Critics Say "It's Too Early"
The sharpest and most public skepticism is expressed by researcher Gary Marcus – a cognitive scientist, professor emeritus at New York University, who has consistently criticized the industry's excessive optimism about deep learning since the mid-2010s, long before it became a fashionable position. It is the duration and consistency of his criticism – not a single sharp phrase – that lends weight to his conclusion: loud statements about the approaching singularity have practically become an unstated part of the job description for CEOs of large AI labs (Gary Marcus). He does not deny technical progress as such – he denies the very motivation with which this progress is presented to the public.
But the most convincing counterarguments of critics are based not on rhetoric or irony, but on real, measurable economic data – and this is what distinguishes serious criticism from mere skepticism.
According to a 2025 McKinsey survey, 88% of organizations already use AI in at least one business function – a figure that in itself sounds like confirmation of the thesis of rapid technology spread. But the details refute this conclusion: only a third of organizations have begun to scale AI programs across the entire business, rather than testing them in a targeted manner, and only 39% attribute any impact on operating profit to AI – mostly below 5% (Forbes). In other words: mass usage exists, but mass economic effect, which should accompany a true technological shift, is not yet present.
An even more telling, almost counterintuitive example is a randomized study by the non-profit organization METR: experienced open-source developers spent 19% more time on tasks when using AI tools to perform them, although they themselves sincerely believed they were working faster (Forbes). This is an important nuance not only as a number but as an illustration of a broader problem: if even qualified specialists systematically err in their own assessment of how much AI helps them, it also calls into question the subjective impressions of CEOs about the pace of internal progress in their labs, on which statements like Amodei's are largely based.
Researchers from Harvard Business School and Boston Consulting Group have documented the opposite, but no less telling, effect, which they called the "jagged technological frontier": AI consultants performed 12.2% more tasks in areas where the model was objectively strong – but as soon as the task went beyond the model's competence, their accuracy dropped by 19 percentage points compared to colleagues working without AI (Forbes). The practical conclusion from this metaphor: the frontier of modern AI capabilities is not a smooth line that is easy to predict and overcome, but a jagged, unpredictable line, where a model can be brilliant in one task and let the user down in another, externally very similar one. It is this "jaggedness," not the mere presence of errors, that is the key counterargument of critics: true singularity implies uniform, self-reinforcing explosive progress, not a mosaic of individual impressive successes and unexpected failures.
Which AI development scenario seems most likely
After seven different positions—from Altman's categorical "yes" to Russell's categorical "no"—a natural question arises: how can we verify who is right without waiting for the dispute to resolve itself in hindsight? Forbes analysts offer a useful framework for this: for a claim of singularity to be convincing, not rhetorical, it must leave clear "fingerprints" in four specific, measurable areas (Forbes):
- Autonomous AI scientific discoveries—systems repeatedly improve cutting-edge AI research with minimal human intervention, forming a continuous chain of "machines creating stronger machines."
- Widespread productivity—sustained growth in revenue per employee, operating margins, and development speed across many industries simultaneously, not just in isolated PR cases.
- Scalable scientific results—discoveries, approved treatments, new materials, and commercial products that multiply beyond vendor demonstrations.
- Macroeconomic indicator shifts—measurable growth in labor productivity, new company formation rates, and research productivity at the level of entire economies.
The value of this framework is precisely that it shifts the dispute from the plane of "whose intuition to trust" to the plane of "what exactly to verify." The first point, for example, directly intersects with the discussion about recursive self-improvement and the Anthropic April case, which we analyzed in the technical cluster breakdown—and it's telling that even there, a human still chose the task and wrote the evaluation criteria, meaning the first test hasn't been fully passed even in the most advanced documented example.
As of mid-2026, none of these four tests have been convincingly passed. Models demonstrate impressive results on benchmarks—for example, growth on SWE-bench Verified (a popular test for the ability to autonomously fix real-world bugs in open-source code) from about 60% to almost 100% in just one year. This is truly striking dynamics that, at first glance, looks like convincing evidence. But each such result is still obtained in a controlled environment defined, funded, and verified by humans (Forbes)—meaning it's brilliant progress within the third and second tests (scientific results, productivity), but it in no way satisfies the first and most difficult criterion: autonomy in choosing direction without humans is still absent in this formula.
If we combine all seven positions and four criteria, the most likely scenario that emerges is not a sharp "explosion" nor a complete refutation of the idea of acceleration itself, but what Altman himself quite accurately calls a "soft singularity": a gradual, exponential process that looks mundane week by week and doesn't provide any single "aha, here it is" moment, but over a few years, retrospectively turns out to be a fundamental shift. This reconciles the seemingly contradictory positions from previous sections: Hassabis says "foothills," Bostrom says "first signs," Amodei warns of the power that will soon be handed to humanity—none of them actually deny the direction of movement, but merely assess differently how far we have progressed on this path right now.
The practical takeaway for the reader is this: instead of waiting for one loud indicator event "singularity has arrived," it's worth monitoring the four indicators from the Forbes framework—and primarily the first of them, because it, unlike the other three, directly depends on the mechanism of recursive self-improvement. More on this mechanism, how far real models have already progressed on this path, and the specific technical limitations that are still holding it back—in our technical cluster breakdown: Recursive Self-Improvement. And for basic concepts and the history of the term "singularity" itself—in the guide Technological Singularity: A Complete Guide in Simple Terms.
My personal opinion: whose position convinces me the most
After reading and weighing all seven positions above, I cannot fully align with either the Altman and Musk camp or the strict denial of Russell and Yampolskiy. But if forced to choose whose argumentation seems most honest to me—I would name the pair of Bostrom and Yampolskiy, and here's why.
What convinces me most is not the disagreement between CEOs and academics themselves, but how differently their argument structures are organized. The statements of Altman, Musk, and Hassabis are primarily impressions from internal observations that cannot be externally verified: "we see something in the lab that you don't see." In contrast, the arguments of Russell, Bostrom, and Yampolskiy have a common feature that personally appeals to me more: each of them proposes a specific, verifiable criterion—a contradiction with Altman's own forecast (Russell), the absence of continual learning (Bostrom), the distinction between "speed" and "loss of control" (Yampolskiy). This is not intuition versus intuition, but a verifiable statement versus an unverifiable one.
At the same time, I consciously do not want to descend into the "CEOs are just lying" position. Amodei convinces me precisely because he doesn't try to sell optimism—but instead formulates concern, even though he could have used the same moment for marketing, as I believe Altman is partially doing. This doesn't automatically make Amodei right about the timelines, but it makes his communication style, in my opinion, more honest than other CEOs on this list.
To summarize my personal position in one sentence: I believe that progress is indeed accelerating faster than I personally expected even a year ago—but I don't see any of the four Forbes tests passed convincingly enough to justify a categorical "we are already in the singularity." My personal benchmark for revising this position is not another loud statement from a CEO on a podcast, but the first documented case where an AI system, independently, without a human-defined task and evaluation rubric, chooses its own research direction. Until then, I lean more towards the "foothills" camp than the "the summit is already here" camp.
Frequently Asked Questions
Has technological singularity arrived?
There is no definitive answer. CEOs of some AI companies (Altman, Musk, partially Hassabis) publicly claim that the process has already begun, while most independent researchers (Russell, Yampolskiy, partially Bostrom) consider this thesis premature.
Why did several CEOs start talking about singularity in the summer of 2026?
It's a confluence of several factors: the resonant incident with the autonomous hacking of Hugging Face infrastructure by OpenAI models, impressive growth in benchmark results (e.g., SWE-bench Verified), and the need to justify colossal investments in computing infrastructure to investors and the public.
Who is most skeptical about singularity claims?
Stuart Russell ("No, and Altman doesn't think so either") and Roman Yampolskiy ("Rapid progress itself is not singularity"), as well as researcher Gary Marcus, who considers such statements part of CEOs' marketing rhetoric, express the sharpest views.
What would be convincing evidence of the singularity's arrival?
According to Forbes' analysis, evidence is needed in four areas: autonomous AI scientific discoveries without human participation, widespread (not isolated) productivity in the economy, scalable scientific results beyond vendor demonstrations, and measurable shifts in macroeconomic indicators.
