Since the release of the artificial intelligence chatbot ChatGPT late last year, users have marveled at the model’s ability to write poetry, crack jokes and rhyme like American rap legend Snoop Dogg. And Albert Webson, a researcher at Brown University, counts himself among those who believe such AI-driven technology has the potential to vastly improve human life by accelerating access to knowledge and automating mundane tasks.

But as people have discovered ways to trick OpenAI’s chatbot into writing password-cracking scripts and constructing incendiary devices, Webson acknowledged that such large language models (LLMs) can also serve