New research from the Wharton School’s Generative  AI Labs shows how large language models can be coaxed into ignoring safety ...