Miggo’s researchers describe the methodology as a form of indirect prompt injection leading to an authorization bypass. The exploit evaded various defenses to detect malicious prompts because the ...