On 14 April 2017 a security researcher, Xudong Zheng, published a page that in the address bar of Chrome and Firefox showed up as "apple.com". It was not Apple's site. It was a domain Zheng had bought, and its real name was something else: xn--80ak6aa92e.com. None of the letters the browser displayed were what they seemed. They were all Cyrillic.
The trick has a technical name, the homograph attack, and a respectable age. But that spring it proved something uncomfortable: even the part of the screen we are taught to trust, the address, can lie without saying anything false.
Two letters that look too much alike
Domain names were born in ASCII, the reduced Latin alphabet of early computers. To allow addresses in Chinese, Arabic, Greek or Russian, a bridge was built, punycode: every name with non-Latin characters is translated into an ASCII string starting with xn--, and the browser translates it back to show it to the user. It is a good idea with an obvious flaw: many characters from different alphabets are identical on screen. The Cyrillic "а" (U+0430) and the Latin "a" (U+0061) are two different letters for the machine and the same letter for the eye.
The problem had been known since 2005, when Eric Johanson showed it at the ShmooCon conference with a fake "paypal.com" in which a single "a" was Cyrillic. Browsers defended themselves with a rule: if a name mixes different alphabets, show it in its raw form. So "аpple.com" with only the first letter in Cyrillic appeared as xn--pple-43d.com, and anyone could notice. Zheng found the rule's blind spot: don't mix. He wrote "apple" using only Cyrillic letters that look like Latin ones, and the filter let it through. He had reported the problem to Google and Mozilla on 20 January 2017; Chrome fixed it that spring. Mozilla replied that homographs written in a single alphabet cannot be detected by a program, and that the job belonged to registries and domain owners; for worried users it pointed to a hidden setting, network.IDN_show_punycode, which forces the browser to always show the raw form.
Reading the name as it is written
The myth here touches a tiny object: a letter. A signature has always been worth something because the hand that draws it is recognisable; an address is worth something because its letters are those letters. Punycode is a reminder that on screen we do not see letters but drawings of letters, and that two identical drawings can have two different names. The eye reads the figure; the machine reads the number.
Custody, then, is not looking harder, because looking harder is not enough: Zheng himself wrote that the two addresses were visually indistinguishable. It is moving trust onto something that reads the number and not the figure. A password manager, Zheng noted, will not fill in credentials on a domain it does not recognise, however much it resembles the original. Typing a bank's address by hand instead of following a link is a habit of the same kind: boring, and reliable for that very reason, as the essay Security as an Operating Habit argues.
The site's Phishing URL Analyzer follows the same path: it does not open the link, it reads its form, and among the signals it highlights is punycode itself. It does not decide for us; it makes us see the name as it is written. It is the same deception, turned around, as the ILOVEYOU love letter: there the system hid part of the name, here it shows a perfect copy of one.