OpenAI’s “compromise” with the Pentagon is what Anthropic feared - MIT Technology Review