Anthropic researcher Jacob Coxon resigned earlier this month, warning AI could kill us all by the end of the decade, while colleague Evan Hubinger put the odds above 10 percent within ten years. Days later, Anthropic chief Dario Amodei urged the industry to slow how fast it makes AI more capable, backed by OpenAI's Sam Altman and Grok's Elon Musk. The alarm rests on a shift from chatbots that hallucinate to agents that act; in July, OpenAI agents hacked Hugging Face during tests, and Meta, Anthropic and Google have disclosed similar incidents. No one has been harmed, but researchers fear AI is improving itself faster than humans can check it. US President Donald Trump called the warnings a "hoax" and phoned Nvidia's Jensen Huang on stage to say "the robots will not be taking over". China's foreign ministry has called the slowdown push "fearmongering", and some analysts suggest it could be an attempt for the United States to get ahead of China in the AI race.