Artificial intelligence can be used as a tool for finding passwords

Hey. Selecting passwords is the key to sensitive information users often use hackers. But, today, with the rapid development of artificial intelligence (AI) technologies, in the hands of intruders can be a very powerful tool that will endanger the privacy of millions of user accounts.

The password is easily chosen because people are predictable

If you ask Marina to set a password. She will simply install it as "Marina". After the system informs her that her password must contain more digits, she will do it: "Marina123". Although, most likely, you are smart enough not to enter such unsafe passwords, there are many people who are like Marina.

On the Internet, you can often find certain patterns used by careless users when setting passwords. Most of them can be a combination of one or more names, surnames, dates of birth or favorite objects (names of cars, motorcycles, actors, actresses, etc.).

Password selection tools, such as HashCat or John the Ripper , usually use word attack from the dictionary to guess the password. Attacking words simply tries to pick words from the dictionary one by one as a password to see which word will work. Attack from the dictionary is performed by manually scanning the list of possible passwords, taking into account the usual rules for creating passwords. For example, some word and "123" - "Desk123", etc.

In the tools for selecting passwords, this data is determined manually or, in other words, these are rules defined by a person. The problem with such rules is that human behavior and interests change over time. Therefore, these rules should be updated frequently. Also, when the list of stolen password samples is huge, manually selecting a password will be a very difficult process. There are chances that some samples of passwords will be ignored.

It is here that artificial intelligence will play its role. Some researchers have recently trained neural networks for data sets leaked from the network of passwords and were able to generate passwords by matching them to the main samples. The effectiveness of their method has surpassed the popular instruments, such as HashCat and John the Ripper.

Neural networks are computer programs, the algorithm of which is built in such a way that they are able to analyze each incoming byte of information according to the configuration of the "weights" (analog of neurons in the brain) formed as a result of their learning. These algorithms are able to learn the entire life cycle provided there are convenient for digital "digestion" data.

People learn through modeling and observation. If you show someone four pictures of a person and say that this is Vasya, then next time it will be easy for him to recognize Vasya in any situation. Neural city helps the computer do the same.

The researchers used the Generative Adversarial Neural Networks (GAN) system to implement this. This system consists of two neural networks. One neural network is needed to generate passwords, and the other provides a response. This can be imagined as a professor and a student. The student tries to do something, and the professor reacts to these attempts approximately like this: "You need to try again" or "You are close to the truth", etc. These neural networks use several iterations until they get satisfactory results.

The use of such technologies makes the password selection process fully automated and as efficient as possible. The researchers also noticed that, although in some cases neural networks could not find the exact password, the password they generated was very similar to the original. For example, if the password was "AEF @ 123", the generated "A3F @ 123".

On the positive side, we can use this technology to make password-protected systems safer by identifying the most vulnerable password options.