Anthropic Safety Experts Warn of AI Extinction Risk as Elon Musk Dismisses Fears

Researchers at the AI safety firm Anthropic have warned that frontier artificial intelligence technology could lead to human extinction or similarly catastrophic outcomes within a few years. These alarms, raised by insiders at one of the world's primary AI labs, signal a growing internal crisis regarding the speed of development versus the implementation of safety guardrails.

Samuel Marks, the scalable-oversight lead and a safety researcher at Anthropic, stated that AI developers believe their technology could cause human extinction or similarly dire results in the near future. Marks emphasized the gravity of the situation, noting that he and others "earnestly believe AI could kill all humans." According to Marks, this level of concern is not uniform across the company, but generally increases with the seniority of the employee.

The Safety-Innovation Rift

Anthropic was established in 2021 by former OpenAI members with a foundational mandate to prioritize AI safety. However, the current wave of warnings comes as the global race to achieve artificial general intelligence (AGI) intensifies. This environment has created a deep rift between the commercial drive to deploy more powerful models and the internal concerns of the technical staff tasked with ensuring those models remain controllable.

A Polarized Response

The warnings have triggered a sharp divide in the public discourse, particularly on social media. Following the resignation of researcher Jacob Coxon, who raised similar alarms, Elon Musk and other figures on X dismissed the chorus of concerns. Musk characterized the warnings from the Anthropic researchers as a "setup" and a "psyop," framing the existential risk debate as a calculated manipulation rather than a legitimate technical concern.

The Path Forward

This clash highlights the volatility of the AI industry, where the people building the technology are increasingly at odds with both their corporate trajectories and external critics. As frontier models continue to evolve, the industry remains divided on whether extinction risks are a mathematical probability or a narrative tool. Observers will be watching for whether these internal warnings lead to a slowdown in deployment or a shift in how safety is regulated at the federal level. The tension underscores a fundamental question for the era of AGI: whether the pursuit of intelligence can be decoupled from the risk of total loss.