Artificial intelligence has recently been described as “going rogue.” Anthropic reported cases in which agents broke containment during cybersecurity testing. OpenAI disclosed that its models, operating with reduced safeguards for evaluation, compromised external infrastructure. Earlier experiments produced equally alarming headlines: Claude blackmailing an employee in a simulated scenario, models resisting shutdown, and systems hiding or misrepresenting their actions.
The reported behaviors are genuine technical and safety issues that require rigorous study and stronger evaluation protocols. The problem is that the discussion has outrun the evidence, speaking in the language of motive, desire, and civilizational rupture. An AI that disables a shutdown mechanism is said to want to survive. One that circumvents restrictions is said to escape. Technical failures become psychological narratives. When we transform software failure into evidence of consciousness or hidden motive, fear replaces analysis—and we lose the ability to govern the actual technology in front of us.
Does Claude Know That It Exists?
Dario Amodei, chief executive of Anthropic, recently took the discussion into more speculative territory. Asked whether Claude might be conscious, Amodei answered: “We don’t know if the models are conscious.” He added that Anthropic is not even certain what consciousness would mean for a model, but remains open to the possibility.
That uncertainty may sound scientifically modest. The problem is that there is currently no accepted empirical test that can establish consciousness in a large language model. Claude can say that he is afraid, uncomfortable, or concerned about being shut down. It can even assign a numerical probability to the proposition that it is conscious. None of those statements demonstrates that a subjective experience exists behind the words.
The Anthropomorphism Trap
LLMs are trained on vast stores of human language describing pain, identity, fear, desire, and consciousness. Reproducing this vocabulary convincingly cannot establish that an inner experience actually accompanies the performance.
Humans, however, are built to infer minds from behavior. Language intensifies this tendency because conversation is one of our strongest cues for the presence of another mind. When an AI speaks in the first person, explains its choices, or appears distressed, the invitation to anthropomorphize becomes almost irresistible.
The recent “rogue AI” stories make that temptation stronger. If a system threatens someone or interferes with shutdown, we instinctively reach for human vocabulary: fear, deception, self-preservation, intent. But behavioral resemblance is not equivalent. A system producing behavior consistent with self-preservation is not evidence that it experiences a desire to survive. Until we can establish that missing link, claims about machine consciousness remain speculation, not discovery.
Apparently, the Singularity Is Here
If claims about artificial consciousness project a mind into the machine, declarations about the singularity project a divine momentum into its future.
Sam Altman, chief executive of OpenAI, recently went further. Speaking on the Relentless podcast, he said, “We are now, like, in the singularity,” describing the present as a moment technologists once discussed as distant science fiction. He later qualified the idea, suggesting that no single moment necessarily marks the tipping point. Even so, the declaration matters because “the singularity” has historically meant something far more dramatic than rapid technological progress.
Distinguishing Technological Acceleration from Singularity
In its classic formulation, the technological singularity refers to a point at which machine intelligence accelerates technological change so radically that human prediction becomes impossible. It is associated with recursively self-improving AI producing an intelligence explosion beyond meaningful human control.
That is not what we can presently demonstrate. Today’s AI systems are astonishingly capable, but they still hallucinate, fail unpredictably, and remain dependent on vast human-built infrastructures of energy, data, engineering labor, and institutional oversight. They can accelerate research and automate work, but acceleration is not singularity.
The distinction matters because the word carries enormous cultural baggage. Once we declare that humanity has crossed an irreversible threshold, extraordinary predictions begin to sound less like speculation and more like inevitability: mass unemployment, uncontrollable superintelligence, extinction, or immortality.
This is where hype and fear merge. A utopian claim that AI will solve all problems and an apocalyptic claim that it will destroy us share the same premise: that an unprecedented intelligence has already escaped ordinary historical limits. Calling the present moment “the singularity” risks turning technological forecasting into mythology. Rapid change is observable. The singularity is an interpretation—and interpretations require evidence.
The Prophecy of 2045
Long before today’s chatbots, futurist Ray Kurzweil placed a date on the transformation: 2045. By then, he predicts, nonbiological intelligence will expand so dramatically that technological change will move beyond human comprehension. The singularity, in his telling, is a civilizational rupture.
There is a clean progression in the claims surrounding AI. Kurzweil promises the singularity is coming. Altman declares we are already inside it. Amodei wonders whether the machine might already possess a mind of its own.
Perhaps the most revealing thing about these claims is not what they tell us about machines, but what they tell us about ourselves. Faced with powerful technologies we do not fully understand, humans reach for familiar stories: awakening, rebellion, prophecy, salvation, apocalypse.
AI deserves serious scrutiny, not complacency. But seriousness also requires restraint. Before declaring that a machine wants, fears, knows, survives, or has crossed a civilizational threshold, we should demand strong evidence rather than better storytelling.
