Chatbots powered by artificial intelligence are remarkable for their ability to generate responses that mimic human conversation.

But cybersecurity experts are warning that anyone could use the software to write deceptive phishing emails or source codes for malware.

One such application that has taken the online world by storm is ChatGPT. The program was released by US startup OpenAI last November.

It can respond to users' queries in clear, logical prose and even write novels or computer code.

Japanese cybersecurity firm, Macnica, tested ChatGPT to gauge the risk for misuse.

Researchers asked the chatbot to write a phishing email designed to lure people to a deceptive website. ChatGPT produced a text in natural Japanese.

The chatbot also generated a ransomware computer code which proved capable of encrypting the hacked files.

ChatGPT is specifically designed not to respond to queries potentially linked to illegal activity. The software refused to answer direct commands to write malware, but it complied when the questions were rephrased.

Experts say underground forums for hackers are already buzzing with discussion of how to utilize the chatbots when carrying out cyberattacks.

Shinogi Shota, who works at Macnica, said "Even people without any basic skills can easily get involved in cybercrime. That's very worrying."

Shinogi said that if overseas scammers can easily produce phishing emails in Japanese, they can expand their targets to Japan. He said "The software could end up lowering the bar for crimes."

Another Japanese expert, who studies the relationship between humans and AI, says people using generative AI software need to make an ethical choice of whether or not to use the program.

Professor Sato Ichiro of the National Institute of Informatics said "We should never forget that it's humans who control AI, not the other way around."