Anthropic has trained Claude chatbot to ‘push back’ against humans — and results could be ‘disastrous,’ top Microsoft executive warns