While the AI itself wasn’t weaponized, the technique raises concerns about AI agents with broad system access.