AI Has Cut the Cost of Attack. Defense Is Still Expensive.
A friend's relative lost several hundred thousand yuan to an AI investment scam. The dashboard looked real. Across Japan, China and beyond, AI has cut the cost of fraud while defense remains expensive.
Day 1,345 in Japan.
A friend in China told me that a relative lost several hundred thousand yuan to an "AI investment project." The pitch promised high returns to investors who bought shares, with the principal returned in three months. The offer came with a live dashboard displaying real-time project metrics and sales figures.
I spent 10 years working in finance, and I recognized the structure immediately. It was an old-fashioned Ponzi scheme wrapped in an AI narrative. You might assume you would never fall for such a tired trick. The con itself is nothing new, yet the economics behind it have shifted. Building a convincing fake system used to require hiring a technical team and spending time on development, a barrier that deterred casual fraudsters. Now a single AI tool can generate a polished data dashboard in half a day. For an ordinary person without a background in finance, telling the fake from a real operation has become nearly impossible.
The familiar worry about AI is that it will turn against us. The danger arriving first is simpler: bad actors have already seized the advantage. Artificial intelligence has collapsed the cost of building a convincing lie, while the cost of defense has barely moved.
That drop in the cost of deception is especially visible in Japan, where I live now.
Disclosed data breaches in Japan have climbed sharply this year. According to figures compiled by the MM Research Institute, a Japanese research firm that tracks publicly disclosed incidents, cyberattacks led to 454 breaches between January and September 2026. That represents a 49 percent jump from the 304 incidents recorded during the same period in 2025. By Oct. 13, the cumulative total reached 485, surpassing the 432 breaches recorded across all of 2025. Data from Tokyo Shoko Research, a corporate credit research firm, shows the scale of the largest breaches. Across the 14 years from 2012 through 2025, Japan recorded only three breaches involving 10 million or more records. In 2026 alone, it has already suffered three.
Living here, I know most of these names from daily life. Sagawa Express, one of Japan's largest delivery companies, exposed up to 53.23 million records. The travel booking platform skyticket lost about 14.64 million records, and the telecommunications giant KDDI lost about 12.23 million. Even Yakiniku King, a popular barbecue chain, had about 10.78 million records compromised. At Times Car, a major car-sharing service, a breach affected about 6.6 million accounts, including roughly 1.6 million images of identification documents. Japan's National Police Agency reported that in the first half of 2026, suspicious access attempts per IP address averaged more than 13,000 a day, up more than 50 percent from a year earlier.
The cause was not sudden wealth. Japan's traditional shield stopped working. Nobuo Miwa, president of the cybersecurity firm S&J, has a hypothesis: generative AI erased the language barrier. English-speaking markets have long been a crowded battleground for hackers. Japan remained relatively insulated because the language is difficult to master and business customs are distinctive. Foreign attackers used to struggle with those nuances. Now generative AI lets them draft natural Japanese phishing emails and navigate Japanese-language systems for next to nothing.
When that linguistic moat dried up, Japanese networks stood exposed. At Nikkei, Japan's flagship financial newspaper group, about 9,000 emails impersonating employees were sent to internal staff and outside business partners. When IDC Frontier, a Japanese cloud and data center provider, suffered a breach, the single incident affected 495 corporate and government clients. Attackers only need to change a language prompt to cross borders. Japanese companies, from giants to Main Street firms, must upgrade aging IT systems and audit complex supply chains, tasks that demand heavy spending and time. Verizon's 2026 Data Breach Investigations Report shows the same pattern. Exploited vulnerabilities were the initial access route in 31 percent of breaches, up 55 percent from the previous year, while third-party suppliers were involved in 48 percent, up 60 percent.
Japan's experience shows how legacy systems and supply chains struggle against cross-language attacks. China reveals another side of the same shift: fraud has become an industrial assembly line.
On Feb. 26, 2026, China's Supreme People's Court released data on the trajectory of fraud cases. After falling to a low of 52,000 in 2022, the number rebounded to 94,000 in 2024 and stayed at 93,000 in 2025. Court officials said at the briefing that AI face-swaps and deepfakes have been used in fraud. On July 24, 2026, the Ministry of Public Security said AI face-swaps and deepfakes have appeared in telecom fraud, privacy violations, and cases that erode social trust. In the underground economy, operators use automated "traffic-generation scripts" and "promotion gadgets" to screen victims at low cost and high efficiency. Casting a wide net once required human callers reading scripts on the phone. Today off-the-shelf software handles the entire process, from filtering targets to building fake dashboards and swapping faces.
Globally, the speed gap has compressed into minutes.
Data from CrowdStrike's 2026 report shows that, in the attacks it observed, AI-driven attacks rose 89 percent year on year. The company also found that eCrime breakout time, the time it takes an intruder to move from an initial compromised device to other parts of a network, fell to 29 minutes on average. IBM's 2026 Cost of a Data Breach report measures the damage. The global average reached $4.99 million, up 12 percent. When AI was involved, the average rose to $6 million. Breaches tied to "shadow AI," the unauthorized use of AI tools inside a company, rose from 20 percent to 43 percent.
The arithmetic favors the attackers. For them, the marginal cost of finding software flaws, writing foreign-language phishing emails, building fake investment platforms, and filtering victims has fallen near zero, while their speed has compressed to minutes. For defenders, the cost stays high and the pace stays slow, whether the defender is a company patching legacy servers and auditing third-party vendors or an individual updating personal habits. Technology is accelerating while human judgment and older institutions move at their usual pace.
Old habits for spotting scams, such as checking whether a website looks professional, listening for a local accent, or trusting a live video call, are losing their value. Instead of reading the news with growing anxiety, ordinary people can change their own rules for verification. Defending yourself does not require becoming a cybersecurity specialist. It comes down to three habits.
First, slow down whenever money or sensitive data is on the table. Cyberattacks unfold in minutes, and investment scams promise payback in months. Both are designed to close before you have time to think. Deliberately dragging out the pace gives hastily built illusions time to fall apart.
Second, abandon the assumption that seeing is believing. Add a step of verification. A detailed real-time dashboard, an email written in a familiar tone, or even a video of an acquaintance should never be treated as proof on its own. Treat each one as an unverified message from the other party.
Third, verify through an independent channel. If an email asks for a wire transfer or a password reset, close the message and call the person at the number you normally use. If an investment opportunity or a plea for help appears online, meet in person or reach out through a separate messaging app you both use.
The lever in the hands of bad actors has grown longer. Yet if you change the moment of decision from asking whether something looks real to checking it through a second channel, these low-cost attacks lose much of their power over you.
The global economy is weak right now. When an ordinary person stumbles across a bargain, it is not good news. The pie falling from the sky is made of iron.