AI safety is falling behind the pace of development

Coxon warned that the technology could 'kill us all by the end of the decade.' The excerpt does not give a specific date for the departure, his job title, or any comment from Anthropic.

The article argues that AI safety is not keeping pace with development. It states that businesses need continuous monitoring, stronger guardrails and human review of outputs, rather than a one-time test suite at launch. The comparison it draws is between testing performed once, at the point of release, and oversight that continues while a system is in use. The headline frames the gap as AI safety falling behind the pace of development.

The excerpt does not name the businesses the recommendation is addressed to, describe the specific capabilities Coxon judged to be risky, or state whether regulators, other labs or Anthropic responded to his departure. It does not quote anyone other than Coxon, and it gives no data on testing practices, incident counts or safety staffing. It also states no timeline beyond the reference to the end of the decade. As reported, the article presents continuous monitoring, stronger guardrails and human review of outputs as the alternative to a single round of testing at launch.