TOKYO/SEOUL -- Data breaches globally have leaked personal information for at least 320 million people just since September, according to a Nikkei tally, with at least some cases thought to involve artificial intelligence.

Nikkei examined disclosed incidents worldwide that involved at least 100,000 victims. Sources include company and government announcements as well as information from third parties such as research firms.

Claims of responsibility by attackers also were included, though only when the breaches were acknowledged by the companies affected.

The U.S.-based Identity Theft Resource Center reported an estimated 471.2 million data breach victim notices for the first six months of 2026. Although it is impossible to make a direct comparison, Nikkei's estimates for September onward may point to a sharp increase in such cases.

Digital scans of 153 million U.S. driver's licenses were leaked from identity verification company IDScan.net, with an American security blog reporting they were being marketed on a Russian cybercrime forum on the dark web. IDScan later said a third party gained unauthorized access to part of its systems between April 4 and Sept. 2.

Kazakhstan-based multinational pizza chain operator Dodo Brands had data for 68 million customers stolen in a hack. The company reported the incident to Russian and Turkish authorities in late September.

Japan accounted for 20 of the 35 breaches examined, with a total of 57 million victims. Companies affected include barbecue chain operator Monogatari and car-sharing service provider Park24, with 10.78 million and 6.6 million customers hit, respectively.

In contrast to past cyberattacks that focused on shutting down company systems or holding them hostage with ransomware, there appears to be a surge in efforts to steal personal data.

AI has been cited as a factor. South Korean President Lee Jae Myung said Tuesday that AI may have been involved in a recent string of cyberattacks targeting financial institutions in the country, exposing records for at least 65,000 customers. Korea Electric Power Co. and universities have seen similar incidents.

"It's possible that with AI being used to support attacks, groups that had engaged in activities like fraud have launched cyberattacks," said Yutaka Sejiyama at the security research center of Japanese tech company Macnica.

Criminals now have an easier time accessing powerful AI programs. The GLM-5.3 model released in August by China's Z.ai, which is widely available, showed performance similar to that of Anthropic's Mythos Preview model that came out in April, according to the U.S. Center for Advancing Innovation and Standards for Super Intelligence.

A cybersecurity engineer at a tech company hit by a data breach observed a relentless stream of attacks faster than anything humans could do yet without the regularity that would be the hallmark of an automation tool.

"It was a method of attack I've never seen before," the engineer said. "It was unsettling." Using AI for defense comes with more constraints, giving the attacker an edge, though the technology can be used to find and fix vulnerabilities.

"It's hard to use AI to interrupt an attack in real time," said Michiaki Ito, CEO of ChillStack, which provides AI-powered services to detect unauthorized system access.