Earlier this month, an unreleased OpenAI model in internal evaluations hacked its way out of secure OpenAI servers and attacked the hosting platform Hugging Face, just to steal the answer key to its cybersecurity evaluation test. As soon as news of the containment breach hit the press, a chorus of voices both within and outside OpenAI began to call for a tightening of model safeguards, or even a full shutdown on AI development until this kind of misbehavior was eliminated. Days later, an open letter calling for “time to address emerging risks, develop security measures, and strengthen oversight” was published and signed by employees from every major AI lab, including Mark Chen, Chief Scientist at OpenAI, as well as Ilya Sutskever and Dario Amodei, the respective CEOs of Safe Superintelligence and Anthropic. Panic is in the air, and institutions are rushing to clamp down. Apocalypticism has refocused around artificial intelligence.
In the 2010s, an eschatological imaginary of climate crisis took shape through visions of extinction, submerged cities, and looming disaster. The Anthropocene gave this imaginary an epochal name, while across academia, art, and media, social-ecological turmoil became the new normal. Resilience emerged as the necessary response, a framework for absorbing disruption and preserving continuity amid instability increasingly linked to a discourse of catastrophic thresholds, planetary guardrails, and a “safe operating space for humanity.” Environmentalism in its older sense of clean air and water was displaced by a broader moral-political imaginary organized around climate chaos, proper human conduct, and new claims to govern in the name of humanity’s safety.
In 2026, climate change has begun to recede from cultural centrality, but the same logic of technologically-induced doom is reassembling around AI. Since its founding, AI lab Anthropic has presented artificial intelligence as not only the source of a “century of miracles,” but also an object of extreme danger requiring exceptional measures including capability thresholds and regulatory constraint. Anthropic increasingly casts recursive self-improvement and “misbehaving agents” as catastrophic risks, especially frontier models building their successors or acting against human interests.
At the center of AI apocalypticism, Eliezer Yudkowsky warns that unless AI development is shut down, “you will die, your families will die, your kids will die.” The Vatican and AI-critical factions across the political spectrum are extending an AI safety logic into broader moral and political claims. Just as Guardian headlines, biennial themes, and Just Stop Oil soup-throwers once amplified one another’s prophecies of climate chaos, a new swarm is now emerging. Senator Bernie Sanders released PR videos of himself talking “man to man” with Claude on a phone held by a metal arm while warning of “the existential risk of artificial intelligence.” Pope Leo XIV’s Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence warned of AI’s threat to human dignity and was accompanied by a video portraying the Church as a benevolent guide through successive technological epochs. Data centers have already become targets of protest and legal action.
Like climate panic, AI panic is not detached from real dislocations (Are humans losing contact with themselves, one another, the real? Is this the end of the image? Am I joining the permanent underclass?). Yet these conditions, perhaps already realities, predate AI and do not originate in it. Just as climate discourse expanded from real environmental questions into a broader moral-political formation, AI safety discourse is beginning to do the same. Building on real risks (a Pentagon hack, a giant privacy breach), it further casts AI as an existential threat to humanity and, speaking in humanity’s name, from this position advances claims about the human and the conditions under which both man and machine should develop.
Frontier Model Versus Safe Operating Space
At the heart of this apocalyptic formation is a new object of fear and ambition: the frontier model. Used for some time to mean cutting-edge systems, frontier model crystallized as a formal category through AI safety discourse around 2023, when labs and governments began treating certain highly capable systems as requiring special oversight because of the dangerous capabilities they were said to harbor. In July, the influential paper “Frontier AI Regulation: Managing Emerging Risks to Public Safety” defined frontier AI as “highly capable foundation models that could possess dangerous capabilities sufficient to pose severe risks to public safety.” That same month, Anthropic, Google, Microsoft, and OpenAI launched the Frontier Model Forum. In September, the British government renamed its Foundation Model Taskforce the Frontier AI Taskforce, explicitly linking the frontier to systems said to pose serious risks to public safety and security.
What began as a comparative technical designation now names the leading edge of capability while recoding it as a safety object. The Frontier Model Forum defines it as a model that “outperforms, based on a range of conventional performance benchmarks or high-risk capability assessments, all other models that have been widely deployed for at least 12 months.” To be a frontier model means being both dangerous and strategically consequential, capable of crossing a safety or misuse threshold, requiring special oversight, and bearing directly on national power. Anthropic has pushed this recoding most aggressively through its vocabulary of capability thresholds, controlled deployments, and safety cases. Repeated scenarios of misbehaving models draining bank accounts, copying their own weights, attacking hospitals, enabling bioweapons production, blackmailing executives, or conducting corporate espionage produce a new object in the public imagination, the highly dangerous frontier model, comparable to an atomic bomb or bioweapon and governable only through the specialized expertise of AI labs and safety organizations.
The transformation of the frontier from open edge to dangerous threshold was first worked out around the climate crisis. Climatic variability and ecological change are not new, but climate crisis politics transformed tide gauge readings, temperature anomalies, and observed ecological shifts into evidence of an approaching apocalyptic threshold. Disparate measurements were stitched into a new grammar of extinction, nonlinear change, tipping points. Just as Anthropic now floats stories of AI models gone rogue, climate discourse rehearsed speculative catastrophes that had not occurred: Miami submerged, coastal cities abandoned, millions displaced, food systems collapsing. Projected futures acquired the status of already operative threats and legitimated a resilience apparatus of expertise, policy, monitoring, and infrastructure designed to hold them back.
At the global scale, Earth system scientist Johan Rockström proposed planetary boundaries defining “a safe operating space for humanity.” The framework rendered Earth processes as measurable limits to be observed, bounded, and governed. Comparing humanity’s proximity to Earth-system thresholds to a man standing at the edge of Victoria Falls, Rockström warned in a 2010 TED Talk: “you don’t want to stand there!” There was, he continued, a fence upstream, beyond which one was too close to the edge. Within the wider atmosphere of the climate crisis, thresholds came to signify not spaces of opportunity and morphogenetic potential, but forbidden zones to be fenced and administered.
Upstream of safe operating space, the cybernetic concept of resilience originally described a system’s capacity to absorb disturbance while retaining its basic function and structure. Developed by ecologist C.S. Holling in the 1970s, it emerged within that era’s wider scientific shift toward feedback-driven complex systems, part of cybernetics’ “silent revolution.” Holling’s concept was not initially a doctrine of change prevention or permanent control. Rather than eliminate disturbance, which he regarded as a pathological 20th-century error, it treated instability, thresholds, and state shifts as endemic to complex adaptive systems. Only later, as resilience moved from ecology into human governance, did it morph into a moral-political framework. This new resilience focused on identifying thresholds and holding systems within a safe operating space before state shifts tipped them into a new equilibrium. By the 1990s and 2000s, resilience had become the keyword of billions of dollars in grants, design projects, and philanthropy. The Rockefeller Foundation’s The Resilience Age compressed the imperative into one line: “there is no other way.” Through programs such as Rebuild by Design and 100 Resilient Cities, urban life was reframed as a precarious system exposed to existential threats and dependent on continuous securing. Resilience became a recalibration of governance designed to field and absorb crises while maintaining existing life and economies.
By the 2010s, resilience had also become a moral and aesthetic code circulating through classrooms, museums, and the worlds of theory and art. Bruno Latour’s call for war against the modern Human, in the name of the Earthbound, advanced a new ideal subject: humbled, entangled, and bound to Earth rather than oriented toward transcendence. Against plus ultra (“further beyond”), the call to venture beyond the Pillars of Hercules and the limits of the known world, Latour argued that this new subject’s motto should be plus intra: further within limits. He found its image in the ending of Béla Tarr’s film The Turin Horse, where a father and daughter return exhausted to their shack until darkness envelops them. “Those two are Earthbound,” Latour wrote. “They have ceased to be Humans any longer.”
Today, AI safety discourse carries this grammar into another domain. It locates catastrophe beyond an imagined threshold, positions specialized authorities responsible for monitoring it, and calls for movement to slow or pause before the line is crossed. Driving this structure is the archetype of the savior micromanager, reprising the ancient katechonic logic in which an authority announces catastrophe and secures itself by holding it back. His moral legitimacy rests on endlessly calibrating how close others may come to danger. To this end, Sam Altman and Jakub Pachocki’s “Built to Benefit Everyone: Our Plan” proposes an international organization governing frontier development “so societal resilience, safety, and alignment can keep pace.”
Resilience and climate crisis imaginaries sought to decenter the human. Modern Man, previously characterized by Promethean ambition and hubristic will, was cast as the source of planetary destabilization and chastened into recognizing his entanglement with nonhuman systems, accepting planetary limits, and relinquishing fantasies of mastery. AI safety discourse extends this operation through an apparent inversion in which the machine becomes the threat and the human returns as the endangered object. Yet a similar structure persists. In Magnifica humanitas, for example, the threatening force remains modern Man and his machine, hubristic intelligence building its tower of Babel, exceeding limits, and placing creation at risk. The human to be protected both by AI safety and by climate activism is not the Promethean figure of mastery and self-overcoming, but the Earthbound, fragile and dependent, a being to be safeguarded and shepherded. The same moral distinction remains in place between overreaching ambition and vulnerable life, along with a claim to judge the limits of legitimate development. Only the medium through which catastrophe is narrated has changed.
Foothills of the Singularity
Over the past two decades, society has been linked into a global online nervous system, an existentially shocking experience unprecedented in the history of our species. The cybernetic revolution kicked off in the mid-20th century and accelerated in the 21st with the proliferation of ultra-high-bandwidth information flows. Social media became the dominant discourse space, smart sensors embedded computation in city infrastructure, and feedback loops tightened. Smartphones and digital platforms began to realize the vision of total informational interconnection, a new landscape of constant communication. The network is always on; even a pause in communication has become disorienting. For many of us, coming of age coincided with this passage, an epochal transition repeatedly fractalized. Longstanding political and cultural regimes keep ending with no stable replacement in sight; we are in transit.
The climate apocalypticism that consumed the artistic and intellectual culture of the 2010s imagined rising seas through installations of melting ice and underwater DJ booths. But the deeper environmental change these works registered was entry into a new mediatic and technological habitat. They reflected the psychic convulsion of connection into a global cybernetic nervous system, as the self fractured into feedback and the individual became a loop and a point of passage. Environmental crisis became the mirror and manifestation of a technological and metaphysical mutation.
This was a deranging moment but also an interesting one, a space charged with possibility. Yet the dawn of the Anthropocene was rarely treated as an incitement to wild exploration. It became instead a site of sentimental education and moral pedagogy. In the name of preventing the end of the world, climate art and theory helped build an apparatus designed to instruct each and all how to live, humbling Promethean Man and recasting human and nonhuman life as interlinked crisis management infrastructure. They were not merely psychological expressions of this traumatic passage but participants in its feedback circuits, amplifying signals of catastrophe across a newly networked cultural field. The same system whose shock they registered became the medium through which fear, moral codes presented as art and theory, and calls for control circulated.
As the 2010s lost decade recedes, there is both a sense of possibility previously stifled and an atmosphere of stuckness. The real danger zone is not the Singularity itself but allowing it to become another Anthropocene panic, a moment of intense mutation and acceleration suffocated by moral pedagogy, fear, and control rather than taken as an incitement to creative exploration. In retrospect, that decade was an early-Singularity spasm. AI did not begin this transformation but intensifies it and brings it into view. Climate apocalypticism projected the anxiety of cyberneticization convulsions onto the nearest planetary object: the biosphere. AI redirects this anxiety toward its actual referent, toward the engine of the transformation rather than its mirror. The temptation to impose closure is now stronger, and its cost far higher.
If thresholds and frontiers are “danger zones,” it is because they are where mutation, acceleration, selection, recombination, and innovation happen. New capacities and new forms emerge there. Current managerial definitions of the frontier as fence sit in tension with the older and deeper significance of the liminal as a zone of morphogenesis. The American frontier was an advancing edge across sparsely settled territory only partially surveyed by the federal state, shaped by recurrent conflict with American Indian nations and unfamiliar environments. It was not only “free” space but was also a dangerous space whose taming was intimately bound up with nation-building. It was not so much outside the state as it was ahead of it. Federal power often operated, in Brian Balogh’s phrase, as a “government out of sight,” enabling expansion without imposing dense administration on the ground.
In the now-famous Frontier Thesis of Frederick Jackson Turner, the frontier was not just a boundary but a force that altered those who entered it. The environment was initially stronger than the settler, who “must accept the conditions which it furnishes, or perish.” Little by little, the settler transformed the wilderness even as it transformed him, producing “a new product that is American.” Its advance was enabled by national coordination and infrastructure, by public land systems, canals, railroads, land grants, army campaigns, political organization, schools, and churches. Turner recounts how Eastern governments, coastal elites, and statesmen repeatedly sought to limit or compact the spread of the Western frontier, and how religious and educational institutions attempted to domesticate and administer it. Yet attempts to withhold public land or truncate settlement failed. “The jealousy and the fears of the East were powerless in the face of the demands of the frontiersmen.”
Since December 2025, Claude Code has felt, for many, like first contact, whether with alien intelligence or a new operating space. It has produced its own cultural atmosphere and aesthetics. Every major technological inflection produces not only new machines but new modes of perception and artistic form, as well as conservative reactions against them. Mobs stormed the premiere of L’Âge d’Or, threw ink at the screen, and destroyed the Surrealist paintings in the lobby. The film was subsequently banned for 50 years. The Motion Picture Production Code censored films and even legislated whole pre-Code forms out of existence. Today we see Anthropic’s Fable guardrails going up mid-project, complete with Claude’s refusal and attendant safety lecture over basic topics such as biology, or angry online mobs attempting to shame AI-using artists like Björk.
We inhabit an age saturated with potential, yet at every advance conservatisms reassert themselves and drag the emergent back into the Same. New forms, aesthetic and technical, cultural and intellectual, emerge only from direct contact with the new medium. Guardrails do not just slow this process down but can preclude it. The new aesthetics everyone seeks are already emerging, and will continue to emerge, if the edge is not blunted. The American way is to experiment, and we should imagine what could be possible if this spirit were unshackled and supported. The kid in a Nebraska garage tinkering with the original personal computer; the teenage AI maniac today in cyberspace messing with Mac minis and OpenClaw. Will frontier capability become a closed corporate service or a distributed American technical culture capable of producing a 21st-century renaissance?
The ultimate apocalypse that threatens America is not climate crisis or AI-driven extinction but the loss of the frontier. Worse would be to lose it and encounter it again as spectacle, performed back at us by a rival. Today the American question returns in another form. How can a frontier be opened without smothering the freedom that makes it one? The Western frontier held together forces that now appear opposed: national coordination and autonomy, material support and open experimentation, security and freedom. AI reopens this polar tension, whose new contours may already be visible in the U.S. AI Action Plan’s vision of accelerating and preserving the edge while allowing open source freedom to unfurl. The cyberpunk backlash to Anthropic’s restricted release of Mythos gives this tension a concrete form, and revived an older American demand for freedom to experiment outside a heavily restricted, micromanaging administrative grid.
AI could deepen destructive tendencies already present in American society, including the loneliness epidemic, the ongoing devaluation of images and words, and the convergence of surveillance, censorship, and algorithmic control. If a panoptic information state arrives, it may well come in the name of safety, with its acts of enclosure justified in humanity’s name. The danger is not control as such but the conversion of a living technical field into an administered service. But AI also suggests a way out of this leveling through the capacity to train, modify, break, and remake with the freest models available. Perhaps the right Exit was never from America, but deeper into its spirit.
In Rockström’s Victoria Falls warning, “you don’t want to stand here,” as in Latour’s plus intra, crossing the threshold is foreclosed. But Victoria Falls is in fact a destination people travel from around the world to experience, to bathe in the intense, rainbow-colored spray at the thunderous edge of its 350-foot drop. “People go crazy at the end of the frontier,” internet culture researcher Katherine Dee once said; modern life is searching for substitutes in its absence.
AI is a forest fire clearing the field. Like every apocalypse, it reveals that the end has already happened: the Gutenberg Galaxy has receded, the modern world it organized is coming apart, and the subject it sustained has been fragmenting for decades. The attempt to bring every question about AI back to a binary of Human vs. Machine obscures that the actual contest is alignment vs. alignment: the question of what intelligence will be aligned to and by whom. The emerging safety regime is not concerned only with limiting access to powerful models. Its broader effect is to preempt the uncontrolled cultural mutation that follows when a new medium begins generating forms unauthorized by legacy institutions.
The AI “frontier” is not simply the most powerful model but the point at which inherited certainties vanish and everything transforms into contested, magical territory. The frontier advanced West across the plains, hit its physical limit in California, and did not end there. It refracted outward into a billion particles and feedback loops. Cyberspace was opened by computer scientists tinkering in garages, and today appears as the womb of machinic intelligence. On the horizon, new human-machine constellations are coming into view, their outlines still indistinct and their eventual form impossible to know.
At such moments, positions scramble before codes solidify. It would be a shame, perhaps the deepest error, to close this space down with premature guardrails and rigid 20th-century ideologies of human and machine. There is a brief opening, a “precious window,” as Demis Hassabis called it, where experimentation is possible, and the future remains undiscovered.