Former Anthropic Researcher Warns of AI Existential Risk by 2030
Jacob Coxon's resignation from Anthropic highlights the dangerous race toward self-improving superintelligence.
The resignation of a prominent AI researcher has reignited fears that the pursuit of artificial superintelligence is outpacing human safety measures. Jacob Coxon, a Cambridge-trained researcher who previously worked at both OpenAI and Anthropic, quit his position in September 2026 to warn that the industry is gambling with human existence.
Coxon claims that leading AI laboratories are racing toward the creation of self-improving superintelligence—a threshold where systems can enhance their own capabilities without human intervention. According to Coxon, some developers within these labs believe this trajectory could lead to technology capable of killing humans by 2030. He described the current industry climate as one where developers are "racing straight to self-improving superintelligence and gambling with our lives."
The Superalignment Struggle
This warning arrives amid a broader internal struggle within the AI industry known as "superalignment": the challenge of ensuring that an intelligence far greater than our own remains under human control. Coxon's concerns were not isolated; they were echoed by other Anthropic employees, including Samuel Marks and Evan Hubinger. The debate centers on whether the commercial pressure to dominate the AI market is overriding the ethical necessity to implement rigorous safety frameworks before a superintelligent system is realized.
Industry and Public Tension
While the warnings are existential, industry leadership has acknowledged varying levels of risk. Anthropic CEO Dario Amodei has pointed to more immediate dangers, such as the potential for AI to generate individualized propaganda, breach global computer systems, and enable mass surveillance. Similarly, OpenAI CEO Sam Altman has suggested that the pace of AI development may need to be adjusted to allow society sufficient time to adapt to these emerging capabilities.
These high-level corporate warnings are now filtering down to the general public, shifting from internal memos to local news coverage. In Charlotte, North Carolina, residents interviewed by WBTV expressed a mixture of skepticism toward the perceived hyperbole and genuine anxiety over the lack of public discourse regarding safety. Some locals highlighted the risks of over-reliance on AI and the lack of transparency surrounding the deployment of hyperscale data centers.
The Path Forward
As the race for superintelligence accelerates, the primary point of contention remains the timeline for safety implementation. Critics argue that once a system becomes self-improving, it may be too late to install safeguards. Observers are now watching to see if regulatory bodies or the labs themselves will implement a "slow-down" or if the competitive drive for dominance will continue to prioritize speed over existential security.