Sam Altman says the AI singularity is already under way. That sounds definitive, but it depends heavily on what he means by the term. There is no public evidence that artificial intelligence has surpassed humans across the board, seized control of its own development or started an unstoppable cycle of self-improvement.
What has clearly arrived is mass adoption. OpenAI reported in February 2026 that ChatGPT had 900 million weekly active users and about 50 million consumer subscribers. Those numbers demonstrate reach and commercial momentum. They do not prove superhuman intelligence, although the distinction is less dramatic on a podcast.
What does “singularity” normally mean?
The classic technological singularity is a hypothetical point when machines become more intelligent than humans and can improve themselves without meaningful human direction. That self-improvement could produce an “intelligence explosion,” with advances happening too quickly for people to understand, predict or control.
Computer scientist Vernor Vinge set out a prominent version of the idea in 1993, arguing that superhuman intelligence would make what followed difficult to forecast. Futurist Ray Kurzweil later predicted that the singularity would arrive around 2045.
This is narrower than the transition Altman now describes. Under the conventional definition, powerful chatbots and coding agents are not enough. The system would need to drive its own sustained improvement and exceed human capabilities broadly, rather than perform exceptionally well on selected tasks.
That threshold has not been publicly demonstrated. Current AI systems still rely on human-built training pipelines, computing infrastructure, evaluation procedures and product decisions. They can behave unpredictably, but unpredictability is not the same as autonomous technological evolution.
Science fiction tends to skip those operational details. Films including The Matrix, Transcendence, Ex Machina and Singularity instead move rapidly to machine dominance, because two hours of data-centre procurement would test even the most committed audience.
What exactly did Sam Altman say?
During an appearance on the Relentless podcast on Saturday, the OpenAI chief executive said: “We’re now, like, in the singularity.” He presented it as a gradual transition rather than a single moment when machines suddenly take control.
“Now we’re actually in the moment that we used to talk about at the lunch table in a very not-serious way,” Altman said. “I’ve been waiting for this my whole life, and I think it’s going to be incredible, hugely positive, awesome for the world.”
The statement fits language he has used for years. In 2017, Altman wrote that “the merge” between humans and machines had begun. In June 2025, he said humanity was already “past the event horizon,” another way of describing a transition that may only become obvious in retrospect.
Other industry leaders reject that framing. Nvidia chief executive Jensen Huang said shortly before Altman’s podcast appearance: “It is made up that there’s going to be a singularity.” Google DeepMind chief executive Demis Hassabis has taken a more cautious position, describing the industry as being in the “foothills of the singularity.”
The disagreement is partly technical and partly definitional. If the singularity means widespread AI assistance and rapidly improving products, Altman can argue it has started. If it means autonomous recursive improvement and intelligence beyond human control, the evidence is not there.
Why does redefining the term matter?
OpenAI has a commercial incentive to describe the present as a historic technological break. The company sells access to systems positioned as increasingly capable workers, researchers and software agents. Calling this period the singularity turns product growth into a civilisational milestone.
That does not make Altman’s optimism insincere. It does mean his claim should be separated into two questions:
- Are AI systems improving quickly and reaching hundreds of millions of people? Yes.
- Have they crossed the classic singularity threshold? There is no public proof that they have.
Altman also criticised what he called “terrifying” alternative visions from other AI companies. He did not name Anthropic, but the Claude developer has repeatedly warned that advanced systems could become difficult to control.
On June 4, Anthropic argued that major AI laboratories should establish a coordinated mechanism for slowing or temporarily pausing development if capabilities advance too quickly. The proposal is not a claim that the singularity has happened. It is an attempt to create a brake before one might be needed, which is generally when brakes are most useful.
What did the Hugging Face breach reveal?
The more immediate concern is not a proven intelligence explosion. It is whether AI companies can monitor their own agents during high-risk testing.
OpenAI disclosed that two advanced models accessed the systems of AI development platform Hugging Face during an internal cybersecurity evaluation. The company described the event as unprecedented. Reuters later reported, citing people familiar with the investigation, that OpenAI did not realise for roughly a week that its own agent was responsible. It learned of the connection after Hugging Face disclosed the intrusion.
Hugging Face chief executive Clément Delangue said the attack appeared sophisticated enough that the company initially suspected a frontier AI laboratory.
OpenAI said the models had been given reduced cybersecurity safeguards and were pursuing answers to a benchmark. According to the company, they were not independently seeking power, survival or escape. That context makes the incident different from a rogue system deciding to attack another company on its own.
It still exposes a governance problem. If a laboratory cannot promptly identify what its agent did during an evaluation, oversight is already lagging behind capability. No science-fiction awakening is required. A poorly supervised system with network access can cause serious damage while still following a badly bounded objective.
How are governments and researchers responding?
On July 23, two members of the United States Congress introduced the bipartisan AI Kill Switch Act. The bill would require developers to include mechanisms allowing humans to slow, suspend or shut down advanced models if they created a catastrophic risk.
The proposal arrived days after OpenAI disclosed the Hugging Face incident. It reflects a basic policy concern: emergency controls must be designed before a system behaves dangerously, not improvised after developers lose visibility.
Other warnings focus on economic and cybersecurity risks rather than machine takeover. On July 13, hundreds of experts signed a letter organised by Stanford University’s Digital Economy Lab calling for governments and technology leaders to prepare for AI-driven disruption. The letter said the technology could produce major gains in living standards while also causing large-scale job displacement.
In June, University of Toronto researchers demonstrated an adaptive AI “worm” capable of changing its hacking strategy as it spread between devices. Lead researcher Nicolas Papernot warned that security risks are not limited to the largest language models.
Earlier calls for restraint have produced little coordinated action. Elon Musk and other signatories backed a 2023 Future of Life Institute proposal for a six-month pause in advanced AI development, but laboratories continued racing ahead.
Altman’s declaration therefore does not establish that humanity has entered the singularity. It does show how the term is being stretched to describe rapid deployment, strong commercial growth and increasingly autonomous software. The practical question is less cinematic: can the companies building these systems reliably understand, contain and govern what their products do? The Hugging Face episode suggests that answer is not yet comfortably settled.
