Simpleform Research Paper Accepted at the 40th Annual Conference of the Japanese Society for Artificial Intelligence

The R&D team at Simpleform Inc. (Meguro-ku, Tokyo; CEO: Shota Tashiro) has had their research paper, "Evaluation of Anti-Money Laundering Detection Accuracy Considering Information Asymmetry: Quantitative Evaluation Using Simulation Data," accepted for the 40th Annual Conference of the Japanese Society for Artificial Intelligence (JSAI), beginning June 8, 2026. The paper will be presented as a poster and subsequently published on the JSAI website.

The paper reveals the following key findings:

- Quantitative analysis using simulation data shows a significant gap in anti-money laundering (AML) detection accuracy between an ideal condition where all transactions across banks are observable (Multi-Bank View) and a realistic condition limited to a single bank's perspective (Single-Bank View).
- Even models performing well under ideal conditions show a major drop in accuracy when limited to a single bank's perspective.
- High-frequency "hub accounts" are structurally prone to information gaps, making it difficult for single banks to grasp the full scope of activities.
- Redesigning models based on local account-level features improved detection accuracy in some cases, though not fundamentally.

Background

Many AML studies are designed on the premise of comprehensive observability of inter-bank transactions. However, in practice, banks are limited to their own data. Simpleform provides products and professional services to support corporate screening operations, aiming for a world where all corporations are connected fairly. Through their work, they observed that risks invisible to single institutions become apparent when cross-sectional information is combined.

Results and Recommendations

Being limited to a single bank's view leads to a significant decrease in AML detection accuracy compared to ideal conditions, especially for high-frequency hub accounts. This accuracy gap is structurally inevitable and cannot be solved by individual bank efforts alone.

The study concludes that improving detection accuracy requires:

- Constructing privacy-conscious inter-bank data sharing platforms.
- Aggregation and utilization of customer/transaction attribute information.

These initiatives not only improve individual bank detection accuracy but also expand the total observable information for the entire financial industry, leading to higher-precision AML detection. Simpleform intends to accelerate AI utilization in the AML sector, including disclosing analytical methods and distributing research data to benefit the industry.