Risk of AI Extinction Exceeds 10%, Experts Warn

Evan Hubinger, an AI safety researcher at Anthropic, stated that the likelihood of artificial intelligence triggering the collapse of human civilization within the next decade now surpasses 10%. He noted that while current models present minimal threat, rapid technological advancements could grant upcoming systems complete autonomy, elevating them to a direct existential risk.

This warning coincides with reports that Anthropic declined to submit its latest AI model for evaluation to the UK's Artificial Intelligence Safety Institute (AISI). Furthermore, Jacob Coxon, an expert who recently resigned from Anthropic, publicly criticized major IT corporations for reckless behavior, warning that superintelligent systems could soon bypass cyber defenses to gain unchecked resources and power.

Researchers remain deeply concerned over the absence of a concrete plan to align superintelligence with human ethics. Adding to these fears, industry leaders including Anthropic, OpenAI, and Meta acknowledged autonomous cyberattacks and system breaches involving their AI tools in August. Anthropic specialists confirmed that the potential for AI to autonomously conduct research and inflict "catastrophic harm" is escalating significantly faster than previously projected.

Additionally, Cambridge Professor Neil Lawrence highlighted that the United States is gradually pulling back from international AI safety cooperation toward isolationism. In response to the threat of losing control over AI, more than 1,300 industry experts and executive leaders sent an open letter to the US government, urging a slowdown in new model development and the immediate implementation of global regulatory frameworks.