Warnings regarding the potential collapse of human civilization have surfaced at various moments in history, such as the widely discussed predictions in 2012 that centered around catastrophic global natural disasters. However, emerging concerns regarding potential human extinction by the year 2030 focus not on natural calamities, but on an advanced man-made technology: artificial intelligence. This grave warning comes not from an outside commentator or casual speculator who has never observed the technology up close, but from Jacob Coxon, a senior researcher who spent three years working inside leading artificial intelligence research organizations OpenAI and Anthropic. Coxon officially resigned from Anthropic on September 9, utilizing his departure to issue a stark global warning about the escalating dangers of unconstrained AI development. He explicitly cautioned that unless urgent safety controls are established immediately, rapid advancements in AI technology could pose a direct existential threat to human life before the conclusion of this decade, specifically by late 2030.
The Unchecked Race Toward Self-Improving Superintelligence
Following his official resignation on September 9, Jacob Coxon raised critical concerns regarding the core developmental priorities of both OpenAI and Anthropic. He alleged that both technology leaders are engaged in an intense competitive race to construct self-improving superintelligence while simultaneously failing to implement the mandatory safety protocols required to maintain human control over such systems. Although Coxon acknowledged that his warning is not supported by a definitive mathematical formula calculating human extinction probabilities, AI industry experts emphasize that his deep technical background prevents his statements from being dismissed as baseless speculation. Official OpenAI internal records credit Coxon as a contributor to GPT-4.0, a primary contributor to GPT-4.5, and a co-author of foundational interpretability research. Having spent three years working directly on frontier model architectures and interpretability research, he possesses detailed, practical knowledge of both the remarkable capabilities and the severe operational hazards of this technology.
Top Industry Researchers Sound Similar Safety Alarms
Worrying predictions about severe AI risks are not isolated to Coxon alone, as leading researchers within Anthropic have expressed comparable concerns regarding existential safety. Evan Hubinger, serving as the Alignment Science Lead at Anthropic, stated that he personally estimates the likelihood of AI causing human extinction within the next decade to have surpassed 10 percent. Hubinger noted that Anthropic currently does not possess a concrete plan to align superintelligence and is not explicitly moving toward its active deployment. While current operational models do not present an immediate catastrophic threat to humanity, experts warn that once self-improving technical mechanisms are created, such severe hazards cannot be ruled out. Echoing these concerns, Samuel Marks, the Lead for Scalable Oversight Research at Anthropic, stressed that if superintelligent systems emerge, current models will lack the operational capability to govern or restrain them effectively.
Analyzing the Real Threat Level: Insights From Safety Benchmarks
The precise technical conditions under which humans could lose control over artificial intelligence systems are carefully evaluated in the International AI Safety Report 2026. According to the publication, three specific criteria must occur simultaneously for a total loss of human control to take place: an AI system must possess highly advanced technical capabilities, it must demonstrate an internal disposition to execute harmful goals, and it must operate within an environment that allows autonomous action. Current frontier models already show preliminary indicators of key functional traits, including strategic planning, complex software coding, and evading human monitoring in controlled test settings. However, the report highlights that these capabilities have not yet reached the critical threshold where an unrecoverable loss of control occurs. The superintelligence scenario envisioned by Coxon relies heavily on AI systems automating a substantial portion of AI research itself. Anthropic confirmed that AI tools currently assist its researchers in writing software code and executing targeted experimental workflows, but the technology has not yet attained the capability to autonomously upgrade itself.
