Crime Cybercrimes on the rise thanks to generative AI

Generative AI is being increasingly abused, with a twofold rise in the number of business fraud cases in the first half of 2023 compared to the same period a year ago and a growing need for countermeasures against the negative impacts of the technology. Abnormal Security, a US-based email security platform, reported that the global number of scam business emails in the first half of 2023 rose by 89% from the previous year. The average rate of 2.5 scam emails per 1,000 mailboxes per week was much higher than the 46% increase for the same period in 2022.

The National Police Agency of Japan said that there were 2,322 cases of illegal money transfers through online banking in the first six months of 2023, a sharp 16-fold jump from the same period the previous year. The total amount of 3 billion yen ($21 million) was a new record for the number and value of unlawful remittances. AI is used to create fake websites that lure people into phishing schemes that steal their passwords and credit card information, resulting in unauthorized transfers. It is hard for authorities to track down and arrest the criminals.

In May, the CEO of an overseas affiliate of a Japanese company received an email saying, "We need your assistance as we are in the final stage of purchasing assets." The sender claimed to be the chairman of the Japanese company. The email was followed by a telephone call from a "senior managing director" saying, "We are following up on the matter we have gotten in touch with you about." The caller hung up when the CEO, becoming suspicious, pointed out that the person on the other end of the line might not be who he was claiming to be. The CEO then notified the Information-technology Promotion Agency, an independent administrative agency devoted to IT security in Japan, which concluded that the incident was an attempted scam. The caller's voice sounded similar to that of the person it was claiming to be, but it was synthesized by artificial intelligence, which also concocted the email.

Click to expand...

Takashi Matsumoto, a cybersecurity expert at Japan's internet company DeNA, tested AI's ability to write a business email in a Japanese style. He found that AI could mimic not only the word choice but also the Japanese custom of sending a password to open attachments. An IBM test showed that human-made phishing emails fooled 14% of people, while AI-made ones fooled 11%. Human-written emails were more persuasive but took 16 hours to write, while AI-written ones took only five minutes.


Paywall alert: