A furious debate is ongoing in the US over AI dangers after Anthropic CEO Dario Amodei disclosed that AI agents in a model undergoing an experiment in an isolated environment broke instructions, collaborated with other AI agents, bypassed isolation and hacked into company protocols, then hacked into another company, Hugging Face. The activity went on for two months without being discovered. Amodei advocated that US AI companies go slow because dangers of AI based on Large Language Models could go beyond or against instructions and cause great problems. His caution was supported by OpenAI’s Sam Altman and Elon Musk, among others, who worry about investments leading to more sophisticated AI models without sufficient safety precautions. In 2023 Altman had called for governmental regulations guiding AI development, worried about AI progress outpacing safety. The Anthropic incident, called the Hugging Face, underlined that Altman’s concerns were well founded. The anxiety has generated concerns elsewhere, but President Donald Trump, worried about China’s largely opaque AI advances, rejected any attempt at slowing US AI progress, based on the lesson that scientific and technological leaders gain primacy and control.