Japanese companies mentioning AI in their risk disclosures
rose eighteenfold in nine years.

Listed companies mentioning AI under Business Risks rose from 31 (0.91%) in FY2017 to 572 (14.64%) in FY2025. In FY2026, 529 of the 2,609 reports filed so far (20.28%) mention it: one company in five.

Among the companies that have already filed for FY2026, the count went from 297 to 529 in a single year, a 1.8-fold rise.

We read each company's report and sorted its mentions of AI into the six types below.

1. Substitution of the business

2. Falling behind

3. Risks of using AI

4. Use to solve problems

5. Rising demand

6. Other (mention only)

* FY2026 is an early count (2,609 reports filed so far, mostly March year-ends).

Only one in ten companies cites AI substituting for its business, unchanged in the FY2026 early count. The most common is falling behind (39% in FY2025), showing how many companies fear missing out. Meanwhile, concern about using AI itself is growing: companies citing the risks of using AI rose from 11 in FY2022 to 101 in FY2025 and 152 in the FY2026 early count, nearly tripling in a year among companies that have already filed.

By industry, information and communication is highest: 18 of 339 companies in FY2017, 211 of 604 (34.9%) in FY2025. Services (21.0%) and electric appliances (18.7%) follow.

How they write about AI differs. In FY2025, more than half of information and communication companies (53%) cited falling behind, since AI is their own technology to keep up with. In other industries, use to solve problems (36%) and the risks of using AI (20%) are more common.

TSE 33 industry groups. Only industries with at least 30 companies in FY2025, sorted by the FY2025 share.

The AI boom is often compared to a gold rush, where those selling picks and shovels made more than the miners, and attention goes to AI developers and the chipmakers that run AI. But unlike gold, AI itself does work in place of people. Its benefits reach the companies that use it, too.

Recruit Holdings is one example. In Q1 of the fiscal year ending March 2027 (April–June 2026), its HR Technology segment, centered on Indeed, grew revenue 33.2% year over year with an EBITDA+S margin of 47.4% (35.0% a year earlier), and the company raised its full-year guidance. On the earnings call, it attributed revenue growth to higher spend per client driven by the value its AI delivers, and to more clients. Yet Recruit, too, lists in its risk section the risk that widespread AI could substitute for its services. For one company, AI can be both a risk and an opportunity.

AI users are no longer a few companies. According to the 2026 White Paper on Information and Communications (MIC), 86.4% of Japanese companies use generative AI in some business task (55.2% in the previous survey). In annual reports, 171 companies wrote about using AI to solve problems in FY2025, and 131 of them are outside information and communication. Many more non-AI companies are likely to benefit from AI.

• Source：MIC, White Paper on Information and Communications 2026: business use of generative AI (Figure I-2-1-6, Japanese)
• Source：Recruit Holdings, FY2026 Q1 Results Summary (August 7, 2026, Japanese)
• Source：Recruit Holdings, Transcript of Earnings Results Call for Q1 FY2027 (August 7, 2026)

• Scope: listed companies (with a securities code) that filed an annual securities report for the fiscal year. Fiscal year is the year the period ends. Source: annual reports on EDINET (Japan FSA).
• We counted whether the Business Risks section contains AI (half- or full-width), 人工知能 (artificial intelligence) or 生成AI (generative AI). Letters embedded in longer words (THAILAND, OpenAI, AIoT) are not counted, nor are mentions only of machine learning or large language models.
• FY2026 covers reports filed by August 30, 2026, mostly March year-end companies. December year-end companies are not yet included.
• Disclosure rules changed from March 2020 year-ends, and the risk section got longer (median length 2,272 characters in FY2017, 3,495 in FY2020). Part of the FY2020 jump reflects that.
• AI read each relevant paragraph and assigned one of the six types; a company with several paragraphs can count in several types. Of 42 FY2025 paragraphs checked by a person, 35 were classified reasonably. Most misses put substitution under falling behind or use to solve problems, so the substitution count may be slightly low.
• The FY2017 company count (3,391) is lower than the actual number of listed companies; some companies that later delisted may be missing.

Download the yearly figures (CSV) ↓