 |
| AI can write fraudulent information better than humans |
Natural language processing technology is developing and reaching a high level. Currently, these technologies are among the most important applications of artificial intelligence.
In a small study, researchers were able to create high-quality phishing email content using the GPT-3 deep learning model and other technologies and services.
This raises concerns that fraudsters will rely on artificial intelligence technologies in the future. This is in addition to inexpensive services and can help them generate fraudulent messages. Phishing emails in particular.
The use of artificial intelligence in fraud attacks
At the Black Hat and Defcon Conference in Las Vegas. A team from the National Agency for Technology in Singapore presented the results of an experiment.
The experiment is based on sending phishing emails. It is one of the most common types of hacker attacks. In addition, the main purpose of these emails is to trick the recipient into clicking on malicious links they contain.
This information was sent to 200 people who were not familiar with the experiment. Some of these letters are written artificially, that is, written by team members. And the other part is written with artificial intelligence.
Sure, the links in these posts aren't malicious links, but they tell the team how many times they've been clicked. As you'd expect, posts written with artificial intelligence received far more fake clicks than posts written by humans.
The team said that building a powerful AI model would cost millions of dollars and that it would be suitable for large companies or governments, not individuals.
Ease of use in the future
On the other hand, scammers may not have to develop a complete system. Instead, they can use pre-defined AI services that can be easily used without programming knowledge.
You can also use open source AI models like OpenAI.
During the experiment, the team not only relied on other available AI services, but also used the OpenAI GPT-3 model, which helps to understand and analyze the target character to facilitate deception.
Despite the excellent results and numbers obtained in this experiment. However, the responsible team clarified that the target sample size was only 200 people and did not disclose the final results. If the same experiment is repeated on a larger sample, the results may differ.
In addition, posts written by real people in the experience may not compare to posts written by professional scammers. But the researchers praised the professionalism of the information written by artificial intelligence.