The platform of stimulated machine learning OpenAI Gym used to increase the secrecy of malicious software. The report on this was read at the DEF CON hacker conference.

Virus auto-developer.

Speaking at the hacker conference DEF CON expert company Endgame Hyrum Anderson (Hyrum Anderson) showed how you can configure the machine learning system so that it will develop malware.

It's not about full-format development of Trojans, but the author of the report managed to use the machine learning system so that it turns the legitimate code into a malicious code, but at the same time antivirus solutions with high probability continued to ignore it.

In general, the algorithm looks like this: the machine learning system introduces small changes into the binary code, runs it through the antivirus engine, analyzes the data received from the antivirus, and, according to them, produces new modifications.

The system, created in Endgame, was trained in this way for 15 hours, producing a total of 100 thousand. Malicious samples. 16% of them turned out to be invisible for anti-virus engines.

Next generation virus writing.

According to Hiram Anderson, this approach allows to deceive including those anti-virus platforms that are themselves based on machine learning systems and AI (artificial intelligence).

Since 2016 year. Some companies involved in the development of such antiviruses began to abuse the definitions of "artificial intelligence" for marketing purposes, arguing that their development is "next-generation antiviruses".

Thus, now "next-generation antiviruses" will have to deal with next-generation viruses.

"As you can see, AI is quite possible to use for malicious purposes, and this, in fact, is a very bleak development of events," said Ksenia Shilak, sales director of SEC Consult. - "Using AI to improve the secretiveness of malware can translate the" arms race "between virus writers and antivirus vendors to a fundamentally new level. Especially if such automatic improvement is put on stream ".

The Endgame system is built on the basis of the OpenAI framework, developed by a non-profit organization headed by Elon Musk,. In April 2016 year. This organization has published a public beta version of the OpenAI Gym, a platform for developing and comparing algorithms for "stimulated learning" of artificial intelligence.

It is on the basis of this platform Endgame and developed their "malicious AI". Its sources are published on GitHub and are available to everyone for use.