📜 ⬆️ ⬇️

Google checks all files uploaded to its services on the basis of hashes - and if necessary, sends data to the police

The recent news caused quite a wide resonance in the West, but it completely passed us by.
In early November, thanks to Google in California, they arrested a child pornography suspect in custody - he uploaded a couple of famous photos to his (closed?) Album on Picasa.

It would seem that nothing unusual, everything agrees with the Terms of Service picas and google:

Child sexual abuse. Google is prohibited from posting images of child sexual abuse. If we receive a notification that someone publishes or distributes such images, we will delete the account of this user and send a report on his actions to the law enforcement agencies.

It is not a good idea. Google has a zero-tolerance policy against child sexual abuse imagery. It’s not a problem. There is a lot of fun in your life.

But the article tells in detail how this “receive notification” is actually implemented.
Google says if you’re looking for your fingerprint.

The company says that it’s not a problem.
')
It is a funeral effect.

In total, every photo uploaded to Picasa (even in a closed album) is checked for a hash match with the base of illegal content.
Given that TOS is the same for other Google services, at least Google Drive is also engaged in similar checks, and possibly Gmail.

In Europe and the United States, this behavior has led to discussions about how acceptable it is and how far Google is ready to go.
Check mail for keywords and file hashes? Maybe check local machines Google Chrome? What can't you do to protect children!

I will add that in the database may meet, and very unexpected results, for example
en.wikipedia.org/wiki/Internet_Watch_Foundation_and_Wikipedia
or hentai - in the states it is legal, but in Australia, Canada, England - no.

Upd.
As a hash function, a non-standard algorithm like md5 or sha-1 is used, but an analogue of PhotoDNA from Microsoft, which can detect a photo even after resizing or changing color.
The program PhotoDNA is able to analyze the image and evaluate its contents, determining whether photos with child porn are placed on the picture. The program has a high data processing speed - for appraisal of photos, the application requires less than five milliseconds, the accuracy of photo appraisal is about 98%. PhotoDNA is able to recognize and evaluate a picture, even if the photo has been edited. At the same time, the number of false positives is also not large - about one per billion analyzed photos.

When creating PhotoDNA technology was used "reliable hashing" (robust hashing). This technology makes it possible, on the basis of a number of features of a digital snapshot, to compare it with pictures from a database of porn images. The set of characteristics - hash tags does not change in the process of editing the picture, unlike other technologies of image hashing. This allows you to define pornography, even on those pictures that have been edited. The photo database, which compares the contents of the images, contains more than 30 million photos and videos. The robust hashing technology can still be compared with the use of human fingerprint analysis.

Source: https://habr.com/ru/post/204760/


All Articles