A paper from Anthropic's Alignment Science team on Alignment Faking in AI large language models