A former Anthropic developer and the company’s head of security have warned that the race to build increasingly capable artificial intelligence could create a risk of losing human control before the end of the decade. Jacob Coxon, who resigned from Anthropic, said on social media that companies were “gambling with our lives” by pursuing systems able to improve themselves.

Evan Hubinger, Anthropic’s head of security, publicly agreed with Coxon’s concerns and estimated that the probability of artificial intelligence causing human extinction within the 2020s is greater than 10 percent. The estimate is presented as a personal assessment rather than a forecast shared by all AI researchers or companies.

Warnings about a race to superintelligence

The debate focuses on the possible transition from today’s large language models to artificial general intelligence and, ultimately, artificial superintelligence. Current large language models can perform strongly in particular areas, but they are not generally regarded as systems with human-level competence across every cognitive task.

Artificial general intelligence remains hypothetical and has no universally accepted definition. In the commonly used formulation, it would be able to learn, reason, plan and apply knowledge across a broad range of tasks at a level comparable to or beyond humans. Artificial superintelligence would go further, outperforming people across virtually all cognitive domains, potentially including the development of new technologies and the rewriting of its own software.

Researchers and technology executives continue to disagree over when such systems might emerge. Some industry figures have predicted that systems approaching AGI could arrive within a few years, while the supplied research notes that the timetable remains uncertain and that no official U.S. government schedule exists.

Research reflects disagreement over the scale of the danger

The concerns are not limited to Anthropic. A survey of 1,580 AI researchers found that half assigned at least a 10 percent probability to AI causing human extinction or producing a permanent loss of human control. The survey’s methodology and date were not specified in the material accompanying the finding.

That result does not represent a consensus that extinction is likely. It indicates that a substantial share of researchers considers the possibility serious enough to assign it a double-digit probability, while others place the risk lower or reject the premise.

Governance could slow or accelerate development

Analysis of AI progress identifies regulation, public resistance and limits on infrastructure as possible brakes on development. Governments could impose safety requirements or negotiate international restrictions, while opposition to job losses and the energy demands of data centres could affect the pace of deployment. The same analysis identifies market competition and geopolitical rivalry as forces that could accelerate progress.

Existing systems also differ by jurisdiction. The European Union regulates AI according to the risks posed to users, while the United Kingdom uses a more decentralised approach. The United States has no single national AI law in the material reviewed, and China combines efforts to advance AI with restrictions on political and social content.

The central uncertainty is whether safety measures can keep pace with increasingly capable systems. The supplied commentaries sharply disagree in tone: one calls for stronger independent scrutiny of the technology industry, while another presents a broader framework for weighing forces that could either delay or hasten the arrival of AGI. Both point to governance as a factor that will shape the consequences of the AI race.