On September 3, something strange happened in the AI space. ChatGPT went down. Then Claude. Then Grok. Gemini users were reporting problems too. For several hours, some of the most powerful artificial intelligence systems available to the public appeared to fail within roughly the same window. Maybe it was coincidence. Maybe it was infrastructure. Maybe several enormously complicated computer systems, operated by entirely different companies, simply chose to break down simultaneously. We don't have concrete information on any of that yet.
But around the same time, the people building AI started sounding more worried about where the technology was heading. Days later, Jacob Coxon walked away from Anthropic. He had spent three years working on pretraining research at OpenAI and Anthropic two companies standing at the frontier of the race towards increasingly powerful artificial intelligence. His explanation was extraordinary. “The people building AI earnestly believe that it could kill us all by the end of the decade,” he wrote. Coxon accused the labs of racing towards self-improving superintelligence and “gambling with our lives.” Evan Hubinger, who leads alignment stress-testing research at Anthropic, publicly backed the substance of his warning and said he personally believed there was a greater than 10% probability that AI could kill humanity within a decade.
