Anthropic was facing an imminent deadline to give the government unrestricted access to its AI model, Claude, which it refused over concerns that the model would be used for mass surveillance and the development or use of autonomous weapons.