Security and safety guardrails in generative AI tools, deployed to prevent malicious uses like prompt injection attacks, can themselves be hacked through a type of prompt injection. Researchers at ...