The alignment problem (the gap between what a frontier model is optimized for and what its operators actually want it to do) sits at the center of every serious argument about catastrophic AI risk. Jacob Coxon, a researcher, attached a specific number to that problem in a post on X, arguing that artificial intelligence carries more than a 10% probability of killing all humans. He named Anthropic and OpenAI directly, characterizing both as companies that are "gambling with our lives."
Other experts weighed in on Coxon's claims. The debate in AI safety research over how to quantify existential risk, and whether published figures carry adequate methodological grounding, remains unresolved.
The problem with attaching a number to extinction risk
Probabilistic estimates for AI-caused human extinction are difficult to evaluate because no standard methodology governs how they are produced. Researchers drawing on formal alignment models will reach different figures than those using expert elicitation, and sensitivity to underlying assumptions tends to be large. That variance matters when a specific figure enters public discourse as a settled claim.
When a number like Coxon's circulates on a platform like X rather than through peer review or a formal technical report, the methodological context that would allow others to evaluate it is typically absent. That absence amplifies the claim's reach while stripping the figure of the scrutiny it would receive in a more structured setting.
Anthropic and OpenAI, the two companies Coxon named, are among the most visible actors in frontier AI development. Both have made safety research a stated priority while continuing to release increasingly capable systems. Coxon's framing treats the two as jointly accountable for a level of risk he considers unacceptable, without distinguishing between their respective approaches.
The "gambling with our lives" characterization is not a technical claim. It is a moral one, and it shifts the burden of proof from whether the 10% figure is correct to whether any nonzero probability of human extinction is compatible with the current pace of development at either company.