Permiso researchers found ChatGPhish, a prompt-injection issue that can cause ChatGPT summaries to display phishing links, ...