NEW YORK: Experts have introduced a new security technique called neuron freezing to make AI chat systems more secure and reliable.
According to the report, AI-based chat systems are not only raising privacy concerns but are also having negative effects on the mental health of some users, where trends such as delusions, psychological stress and severe mental problems have emerged.
In view of these risks, researchers have developed a new method that further strengthens the security system within the chat system and prevents users from escaping security restrictions.
According to experts, the current security system mostly works on a two-pronged approach, in which a question is either fully allowed or completely rejected. However, some users overcome these obstacles by presenting the content in a different way.
The new technique, neuron freezing, identifies specific safety neurons within artificial intelligence and keeps them in a permanently safe state so that the ethical and security boundaries of the system are maintained, no matter how the user tries to bypass the restrictions.
This research was conducted by experts from North Carolina State University in the US, which showed that this method can improve the security consistency of large linguistic models and reduce the possibility of misuse.
The researchers say that this conceptual framework will form the basis for new methods in the future that will enable artificial intelligence to re-check the security of its responses, making the system more secure for users.
