Permiso Security disclosed ChatGPhish, a technique that weaponizes ChatGPT’s Markdown rendering layer to inject malicious links and images into AI-generated summaries, redirecting users to attacker-controlled phishing infrastructure without compromising OpenAI’s backend. Because users implicitly trust content that appears to originate from an AI assistant, this attack surface turns ChatGPT into an unwitting phishing delivery mechanism with no visible indicators of compromise in the conversational interface. For enterprises that have standardized on ChatGPT for research, summarization, or workflow automation, this signals that AI-generated output can no longer be treated as inherently safe, the AI’s credibility becomes the attacker’s most valuable asset.