Gemini parses the invisible directive and appends the attacker’s phishing warning to its summary output. If the user follows the AI-generated notification and follows the attacker’s instructions, this leads to leaked credentials or a phone-based social engineering attack.