Diversify your passwords. This is one of the good practices to adopt to ensure your online accounts. But it may be difficult to find a different password for each of its accounts. Therefore, some Internet users are tempted to use artificial intelligence to generate unique and random words.
However, they are not as safe as they seem, he warns Kaspersky’s cybersecurity company, in a study published on the occasion of World password, this Thursday, May 1.
Because even if Chatgpt, Deepseek or Meta calls you know that a good password includes at least 12 characters, tend to adopt bad practices.
Dictionary words, symbols that return regularly …
Kaspersky made this discovery generating 1,000 passwords with Openai, Deepseek and Meta Chatbots, but also those of other companies. Among them, call and deepseek generated passwords that were not safe enough, because they were formed by dictionary words, but whose letters had been replaced by similar figures (“p@ssw0rd”, “s1mp1el1elon” …). A practice that is also adopted by several Internet users.
If at first glance, the Operai chatbot is a good student, it is far from being. In fact, it has generated passwords that approach random passwords, but the company has observed “schemes that are repeated.” Chatgpt, for example, tends to include “very regularly” the number 9 in them. It appears in almost 800 of the 1,000 passwords generated by the Openai conversational robot. The X symbol appears in each of them.
A problem that “contradicts the random aspect of passwords thus generated,” says Kaspersky. Deepseek and Lllama are also worried, highlighting certain letters more than others. “The ideal random password generator would not favor any letter instead of another, and all the symbols would appear approximately the same number of times,” said the company.
Passwords that can be deciphered rapid
Another problem: Chatbots can forget to include a special character or password figures. This was the case of 26% of those generated by chatgpt 32% per flame and 29% by Deepseek. The last two also have also created passwords of less than 12 characters, Kaspersky said.
Many reasons why the passwords generated by AI do not make the weight against cybercriminals, the company advances. Head of the Data Science team within him, Alexey Antonov has developed an automatic learning algorithm to prove his solidity.
According to this, 88% of those generated by Depseek can decipher in less than an hour with a “sophisticated attack.” A figure that increases to 87% for those created by flame and falls to 33% for those generated by chatgpt. On the other hand, almost 60% of the passwords created by humans can decipher in less than an hour with deciphered tools based on the cloud or modern graphic processors.
Kaspersky advises, like others, using a password administrator to ensure his online accounts instead of having Ia.
Source: BFM TV
