“Following criticism, Anthropic will now make its Fable-5 safeguards for AI development visible, alerting users if a request is refused or rerouted to a less capable model.”