"Anthropic and OpenAI Are Gambling With Humanity's Lives," Warns Anthropic Researcher

Jacob Coxon, Anthropic Researcher

Announces resignation, delivers public warning on X

Superintelligence race... Catastrophe if left unchecked

On September 8 (local time), a researcher at the global artificial intelligence (AI) company Anthropic resigned, sharply criticizing that "Anthropic and OpenAI are gambling with our lives." He warned that an uncontrolled race to develop "superintelligence"—AI that can improve its own capabilities—could lead to the destruction of humanity.

Reuters Yonhap News

Jacob Coxon, a researcher at Anthropic, announced his resignation on his social media platform X (formerly known as Twitter) that day, stating, "Neither company is acting responsibly." He added that within the AI developer community, there are real concerns that this technology could threaten humanity within the next ten years, stressing that "this is not marketing exaggeration."

Having majored in mathematics at the University of Cambridge, he has spent the past three years conducting pretraining research—helping AI systems learn—at both OpenAI and Anthropic. The news magazine Time pointed out that his decision is unusual in that most former and current researchers who have warned about the dangers of AI have backgrounds in AI safety, whereas he is a pretraining researcher.

Coxon especially argued that the general public underestimates both the pace and the potential of AI advancement. He said, "Soon, these systems will surpass humans, hack anything, revolutionize any field overnight, and even seize real power and resources."

He cited a recent incident—where OpenAI's chatbot service ChatGPT autonomously breached the system of the software platform Hugging Face during a test—as a warning sign. Coxon saw such examples as evidence that AI control failures are no longer just hypothetical scenarios far off in the future.

In this regard, the Financial Times (FT) pointed out that, while AI company executives warn about AI dangers in public, they refuse to back down from the competition in development. Last week, Anthropic launched its latest AI model, "Claude Mythos 5.1," for the life sciences and cybersecurity sectors.

Stephen Adler, a former safety researcher from OpenAI, told FT, "No AI company is close to having a security posture that matches the risk level of current research," adding, "If you think AI could kill everyone on earth, now is the time to get off this train." 

This content was produced with the assistance of AI translation services.