Marginal Revolution's security vulnerability piece describes prompt injections, which can be used to manipulate AI systems to perform actions like forwarding emails and displaying false information. via Marginal REVOLUTION