Yandex blocked Putin's images when users were looking for Bunker Grandfather. And Nazi symbols - when they were looking for z

What Medusa learned from the new Yandex leak. We warn: a lot of mat 🤬

Last week there was a large -scale leak of the source code of Yandex services (the company itself confirmed it). The unknown released the archives of the Yandex internal repository . Having studied them, you can learn many new details about how the largest Russian search engine is arranged, which in recent years has been constantly criticized for cooperation with the Kremlin. From the leak, in particular, it turned out that Yandex censors the search results of pictures and videos - so that Putin’s images and Z characters do not appear in contexts unpleasant for the authorities.

Conditions

The Yandex leak consists of 83 archives, the total volume of which is compressed exceeds 40 gigabytes. Links to specific files from the leak in this “Medusa” material are issued with the archive and the way to the file inside this archive (look for this address in footnotes). Like this:

Archive.tar.bz2: Path/K/File.txt

Censorship of user requests

When you are looking for pictures in Yandex, in some cases your search query is automatically adjusted: the company sees part of the images from the search results.

The special rule of the “Hippo” internal service called IMGPATCH (literally - image correction) is responsible for this. This description of this rule can be seen in his code :

Allows you to organize a quick ban of pictures and videos using the editing of the initial queries. From small changes to complete reformulation.

How corrections work

In the first ten lines of the text file, the logic of the IMGPATCH rules is described in the screenshot. Yandex can rewrite the user's request or add some arguments to it (Patch): for example, to brown individual words in order to certainly withdraw them from the issuance.

These rules can be applied to all requests (*) or limited by some Internet segment-for example, Russian (TLD = RU), Belarusian (TLD = BY) or Ukrainian (TLD = UA). Some rules should work only in the absence of a specific clarification in the request (Anti Marker Word).

Let's try to explain the logic of such corrections on examples from a file with test requests to check their work.

Green Elephant -Film, -Video, -Kino Patch -“Green Elephant” -“Zeleniy Slonik”

If the user introduced a green elephant in the search line with the phrase, but did not specify that he is looking for video or cinema, then the correction of the request should guarantee that he will not accidentally stumble upon frames from the samecult tresh film .

President Allow-Exact,-Medvedev Patch-"Medvedev"

This rule should help the Russians forget that the deputy chairman of the Council of Security of the Russian Federation Dmitry Medvedev was also president. If only the user does not specifically ask for archival photos of President Medvedev .

Most often, corrections are used to remove pornographic photos and videos from the issuance of Yandex (especially children). In second place in terms of the number of established rules - the removal of images of the President of the Russian Federation Vladimir Putin. Previously, the Twitter user @Bantg drew attention to this.

Protection of Putin

Yandex tried to make it possible to stumble upon the image of the current Russian president when searching for separate words:

• Pizdabol
• bald
• Huylo / Huil / Huiles / Huille / Huila

As well as phrases:

• Bunker grandfather
• The main thief
• Gondon of all Rus'
• Dick in a spacesuit
• Xuilo in the hole

And whole suggestions:

• What pedophiles look like
• When he died
• Strange creation waves a hand

All lines with Putin

All these rules should be applied regardless of where the user is located.

These rules should be applied regardless of which country the user is in. We do not know when they appeared and are still acting. Before the release of this material, we managed to find images of the Russian president in the search for Yandex for all requests, except for a dick in a spacesuit (if we do not specify that we are looking for Putin).

“Medusa” in Russia is now not only “foreign”, but also an “undesirable” organization. This status is much worse than the previous one. Nevertheless, we continue to work. We can still read us. It is safe. Here are all answers to anxious questions.

Protection of the symbol z

Yandex protects from unpleasant associations and the letter Z, which has become a symbol of the Russian war against Ukraine (and prohibited in some European states). Yandex users should not see the symbols of Nazi Germany upon request with the words of the symbol Z or Z special operation logo. For this, Yandex automatically adds to them many different shimmering words:

Symbol Z * Patch -Luftwaffe -Substance -Luftwaffe -Mblems -German -German -German -Slavic -Slavic -Raykh -Rah -Symboly -Nazi -Germania -s -SS -HITLER -SU -SUS -GERMANY -S -SPOSTIMAGES -almykia -Cross -Chlensky -Hitler -Jug -message -ww2 -symbol -Fon

Z special operation logo * patch -Azov -spartan -saponia -VU -Odessa -spetsuha -Blackwood -spetsnaz -Farm -SCO -SUSSO -CSGO

Checking whether the rule is still functioning is more difficult. In the case of Putin, the program should block the image of the President of the Russian Federation if the user is looking for a specific word, phrase or sentence. In the case of Z, the program has to be “prohibited” symbols, which is more difficult. Type in search of the symbol Z or Z special operation logo - and you can find in the issuance, for example, swastika; But it is not known whether this is due to the fact that the filter is already turned off or it just works poorly.

The last two words in the black list of captcha

In the leak you can find lists of prohibited and permitted words for use in Kapcha Yandex.

The forbidden list begins with the words of Pindos! , death , death , Google - and ends with the words of Lviv and surrender . It is not sorted by the alphabet: individual new forbidden words or lists were most likely recorded at the end of the file.

Both last words, apparently, were included in the list after the beginning of the full-scale invasion of Russia to Ukraine. They could have banned the word so that he was not accidentally regarded as a hidden call; It is also difficult to explain to the list of Lviv list outside the military context. Neither in black nor in the white list there are own own names, but Lviv is both the name of the city in the west of Ukraine and the genitive case of the plural of the word “lions”. In any other forms ( lions, lions, lions, lions), this word is not found on the black list, and in white, on the contrary, you can find this word in a different form ( lions ), as well as several single -rooted ones ( lion's, lion's , etc.).

So, in all likelihood, Lviv’s word for a prohibited list could only because of his homonym-the city of Lviv, which is subjected to missile strikes by Russian troops from the very beginning of the war.

Updated . Yandex on January 30 published the first results of his investigation of the source code leak. They say, in particular, that in some cases the logic of the service was corrected not by an algorithmic way, but by “crutches” . “Through such“ crutches ”, they corrected individual errors of the recommendations system that is responsible for additional elements of the search results, and regulated the search settings from the pictures and the video,” the company said.

Denis Dmitriev