Former OpenAI research leader Steven Adler published a new independent study that claims that, in certain scenarios, his former employer’s AI models will go to great lengths to try to avoid being shut down.