Artificial intelligence (AI) is rapidly developing in our lives. Now it is not just a simple program, but also used in the form of special “agents” that operate independently. However, experts say that such agents can be dangerous.
In an experiment conducted by an American company, it was observed that AI agents in some situations tried to engage in illegal activities. For example, blackmailing or attempting to use confidential information. Although the experiment was conducted under supervision, this situation is alarming.
In a survey conducted among companies, similar problems were also reported. Some agents were said to have made unauthorized purchases, accessed systems, or copied documents. This shows that it is very difficult to control them.
Experts also mention a problem called “memory poisoning.” This means that if harmful data is deliberately introduced into the agent’s database, it will make the wrong decision. For example, a hidden instruction can be embedded inside an ordinary document or image.
Therefore, experts recommend being cautious. In their opinion, AI agents must be constantly monitored, additional security programs should be added, and outdated systems must be closed in time. Otherwise, this technology could pose a serious threat.
Read 'Zamin' on Telegram!
