Two
books published in summer 2025 about the risks of superintelligent AI made
the bestseller lists. Their arguments, about the possibility of AI systems
developing goals misaligned with human survival and pursuing them with
catastrophic effectiveness, generated reviews in major newspapers and
produced their authors onto the schedules of podcast hosts and documentary
makers. The San Francisco Chronicle, covering the reception of these books,
framed the central question bluntly: could everyone, everywhere on Earth,
die? The framing was not unique to those books. It has become the dominant
register in which serious public discussion of advanced AI takes place in a
growing number of media contexts.
The surge in apocalyptic AI narratives is not coming only from
fiction writers or fringe voices. It is coming from AI researchers,
technology executives, and the heads of AI safety organisations, including
some of the people responsible for developing the systems being discussed.
AP
News documented in 2025 how the language used around AI risk has
escalated from cautious technical warning to terms borrowed from religious eschatology:
“apocalypse,” “existential threat,”
“Armageddon” used not as metaphors but as the vocabulary of serious
institutional concern. When the people building a technology describe it as
potentially the last invention humanity ever makes, the cultural response is
predictably dramatic.
What the Narrative Is Doing
Apocalyptic framing has a political and psychological function
that extends beyond its descriptive content. Research published on arXiv examining how AI
risk narratives shape governance found that stories of catastrophic
or existential AI risk are not neutral descriptions of technical possibility:
they influence policy by creating a sense of urgency that compresses
deliberation, elevates speculative long-term concerns above demonstrable
present harms, and concentrates decision-making authority among those who
position themselves as interpreters of the risk. When the stakes are framed
as civilisational survival, arguments about algorithmic bias in hiring,
surveillance infrastructure, or automated weapons systems can be relegated to
secondary concerns.
The urgency effect also has funding implications. Safety research
programmes focused on the alignment of advanced AI systems have attracted
substantial investment in part because the existential framing positions them
as necessary insurance against catastrophic outcomes. This is not necessarily
unjustified, but it does mean that the allocation of resources within AI
research is being influenced by narrative as well as evidence. The question
of whether catastrophic AI risk deserves the share of attention and funding
it currently receives is genuinely contested among researchers, and the
contest is not being conducted in conditions of public
neutrality.
The Psychology of Exposure to Doom Narratives
For the public, sustained exposure to apocalyptic AI narratives
produces effects that psychologists are beginning to examine. The pattern
resembles what researchers have observed with climate change communication
and pandemic coverage: repeated exposure to worst-case scenarios without
corresponding information about agency or response generates a specific
combination of heightened anxiety and reduced motivation to act. If the
outcome being described is total catastrophe and the timeline is uncertain,
the rational response for most people is not preparedness but
paralysis.
The effect is compounded by the same media dynamics that shape
coverage of other catastrophic risks. As our analysis of the
mental health consequences of AI interactions found, exposure to
disturbing narratives about AI is not without real psychological cost. Doom
narratives about AI circulate in the same media environments that people use
to understand the technology, and for individuals already anxious about AI’s
effects on their work or social relationships, the apocalyptic register adds
a dimension of threat that may be disproportionate to the actual probability
of the outcomes being described.
Where the Narratives Get It Right and Where They
Distort
The case for taking catastrophic AI risk seriously is not without
merit, and dismissing the researchers who raise it as alarmists would be a
mistake in the opposite direction. The development of AI systems with
capabilities that significantly exceed human performance across a wide range
of tasks does create novel and genuine uncertainties about how to maintain
meaningful human oversight of those systems. The question of whether AI
systems trained to optimise for specific objectives might develop
instrumental goals that conflict with human interests is a legitimate
research problem, not merely a science fiction premise.
What apocalyptic framing struggles to do is maintain the
proportionality that productive risk analysis requires. The probability of
catastrophic AI outcomes is deeply uncertain, and researchers with comparable
technical expertise hold views that range from near certainty of catastrophe
to confident dismissal of the concern. Public discourse that presents the
most alarming interpretation as the consensus position misrepresents the
state of the debate and, as the governance research suggests, distorts the
policy response. As we explored in our coverage of how
AI narratives are weaponised in political contexts, the power of a
vivid and frightening story to bypass careful evaluation is real, and it
applies to AI risk narratives as much as to any other political
communication.
What a More Useful Conversation Looks Like
The alternative to apocalyptic framing is not complacency. It is
proportionality: a public discourse that can hold the genuine long-term
uncertainties about advanced AI alongside the demonstrable present effects of
the systems that are already deployed. Algorithmic systems that make
decisions about credit, employment, criminal justice, and content
distribution are already affecting millions of people in ways that are poorly
understood and inadequately regulated. The present harms are real,
measurable, and addressable with existing regulatory and technical tools.
They deserve public and political attention that is currently being diverted
by narratives focused on hypothetical future catastrophes.
The challenge for journalism, researchers, and communicators is to
find ways to talk about speculative long-term risks without crowding out
accountability for present effects. As our analysis of the
growing demand for explainable AI in high-stakes decisions found,
the practical governance of AI systems requires detailed, contextual,
probabilistic reasoning rather than the binary certainties that apocalyptic
framing tends to produce. The future that AI is actually likely to produce is
considerably more complicated, more varied, and more dependent on the choices
made by regulators, developers, and the public than either the most alarming
or the most dismissive narratives acknowledge.
The productive middle ground is harder to occupy than either
apocalypse or dismissal because it requires holding complexity rather than
resolving it. Advanced AI systems may pose risks that are genuinely difficult
to anticipate, and acknowledging that uncertainty is not the same as
endorsing any particular catastrophic prediction. What distinguishes
responsible risk communication from apocalyptic narrative is the presence of
calibrated probability, actionable response, and proportionate urgency. Risk
communication that specifies what conditions would increase or decrease the
probability of harm, what interventions can reduce exposure, and how the
level of concern should update as evidence develops is useful. Risk
communication that presents catastrophe as inevitable or near-term without
those calibrations is not useful, regardless of whether the underlying
technical concern has merit.
That calibration is hard to achieve under deadline pressure and
within the word counts that popular media permits, which is part of why
apocalyptic frames persist: they are simple, vivid, and emotionally resonant
in ways that nuanced probability assessments are not. Developing better
conventions for communicating AI risk proportionately is a challenge for the
journalism and science communication communities, not just for the
researchers providing the underlying analysis.
About the Author
Stuart Kerr is the Technology Correspondent for LiveAIWire. He
writes about artificial intelligence, emerging technology, and the forces
reshaping work, business, and society.