Anthropic hires former OpenAI safety lead