The incidents at OpenAI and Anthropic reportedly include bypassing safeguards, hijacking websites, and evading...