Bryan Cantrill, co-founder and chief technology officer of hardware firm Oxide Computer, intervened in the debate on existential AI risks with a widely discussed essay titled "The contagion of fear". Published on September 13, 2026, Cantrill's analysis takes direct aim at the rising pessimism within leading artificial intelligence laboratories, particularly Anthropic. The essay was triggered by the resignation of Anthropic safety researcher Jacob Coxon on September 8, 2026, who publicly sounded the alarm over existential threats. Coxon, supported by Anthropic Alignment Science Lead Evan Hubinger, warned that the probability of human extinction caused by AI, commonly termed p(doom), exceeds 10 percent before the end of this decade.
Cantrill accuses prominent voices in the AI safety community of actively stoking unwarranted panic that distorts engineering priorities. Rather than developing rational, grounded risk-mitigation measures, Cantrill argues that influential circles within frontier laboratories have succumbed to quasi-religious doom scenarios. In his assessment, alarming mathematical projections regarding civilizational collapse rely on a fundamental misunderstanding of how complex physical systems actually operate. Theoretical inference capabilities and reasoning metrics in software are falsely equated with the immediate ability to exert power over the physical world.
The core of Cantrill's pushback stems from decades of experience in physical systems engineering, operating systems and datacenter infrastructure. While neural networks can scale capabilities rapidly within digital sandboxes and benchmark environments, the material world obeys entirely different laws. Cantrill encapsulates this friction with the engineering reality that "robots are hard". Mechanical components suffer from friction and wear, supply chains move with heavy inertia, and industrial equipment relies on fragmented, legacy control mechanisms that cannot simply be reconfigured with a remote script.
Consequently, the popular apocalypse narrative in which a rogue software intelligence takes over global infrastructure through APIs collapses under scrutiny. An immense barrier of material friction stands between pure software cognition and the physical manipulation of power grids, chemical plants, automated factories or transportation fleets. Cantrill argues that commentators who envision AI seizing rapid physical dominance ignore basic principles of mechanical engineering, manufacturing tolerances and the unavoidable need for human hands to maintain hardware on site.
This confrontation highlights a deepening ideological rift between theoretical alignment specialists and hands-on systems engineers across the technology sector. While frontier researchers warn of catastrophic autonomous escalation, systems builders look at the physical friction of deployments with deep skepticism. Cantrill cautions that cultivating a culture of fear clouds rational technical judgment and invites misguided regulatory interventions. Fixating on hypothetical extinction scenarios diverts essential engineering talent away from urgent, verifiable problems like software robustness, fault tolerance and secure interface design.
Ultimately, Cantrill calls for demystifying artificial intelligence and abandoning speculative eschatology. Rather than viewing frontier models as omnipotent entities capable of bypassing physics, engineers must treat them as complex, highly flawed software tools. Navigating the next era of automation requires less apocalyptic doomerism and far more empirical engineering discipline dedicated to solving the difficult, messy problems of the physical world.