Detecting SNS Investment Fraud and Impersonation Ads in Japan
A familiar face, a natural-sounding voice or a bank name is not enough to establish that an investment ad is genuine. Reviewers need a way to find questionable material, preserve the evidence and check the advertiser.
Japan’s 2026 figures
The National Police Agency’s October 2 release reported 7,097 SNS investment-fraud cases and JPY 93.74 billion in losses for January–August 2026, up 53.8% year on year. These are provisional figures and a separate category from SNS romance fraud.
Government and platform measures
In a document dated August 7, 2026, seven Japanese government bodies asked Google, LINE Yahoo, Meta, TikTok and X to strengthen measures including advertiser identity checks and ad transparency. A request is not evidence that every measure has been implemented or verified.
Signals in public ads
Check combinations of signals against official communications and registration records.
| Signal | What to check |
|---|---|
| A famous person or financial institution | Does the official site carry the same offer? |
| Guaranteed returns or limited places | Who is advertising, and what supports the claim? |
| Redirection to LINE or another app | Does the destination match the claimed official contact? |
| Similar ads across accounts | Wording, timing, advertiser information and shared destinations |
Follow the public trail, then mark what is unknown
An investigation may follow the path below. Public posts can reveal an invitation and its destination; later conversations and payments require evidence from an authorized party. A redirect alone does not establish fraud.
- Public ad or postRecord the claimed identity, displayed advertiser, wording, URL and capture time.
- Invitation to a messaging appSave the public invitation link and check whether it matches the organization’s official contact.
- Site or payment requestRecord only material lawfully available to the reviewer. Private chats and transactions may remain unknown.
- Human verificationCompare original evidence with official records, document uncertainty and refer through authorized channels.
AI narrows the queue; people verify
A possible workflow uses text or image similarity to find candidates, then a generative model to organize reasons for review. The model’s explanation must be checked against the original. This describes a design option, not a claim about a deployed model or accuracy.
- Preserve URLs, timestamps, images, displayed advertisers and destinations.
- Group similar content and prioritize items for review.
- Check official statements, business registration and original media.
- Record evidence and decisions; refer or report through authorized channels and retain corrections.
A fictional ad, worked through
Illustrative example, not a real case: an ad uses a fictional bank’s logo, promises guaranteed weekly returns and invites readers to a LINE group.
| What is observed | How to record or check it |
|---|---|
| Logo and claimed endorsement | Save the ad and advertiser information. Check the bank’s official website and contact it through independently verified details. |
| Guaranteed returns and a LINE invitation | Record the exact words and public destination. These are review signals, not proof that a crime occurred. |
| Similar ads from several accounts | Compare wording, images, timing and destinations. Similarity alone does not prove a shared operator. |
| No private conversation or payment evidence | Record those stages as unknown. Keep the item pending until the necessary evidence or an official response is available. |
The review record should name the evidence checked, reviewer, time and outcome. If the advertiser is later verified as legitimate, retain the correction and revise the alert.
Limits and personal information
Public monitoring does not reveal private chats or transaction histories. Legitimate ads and news can share similar wording, and AI explanations may be wrong. Limit personal information, define the purpose and agree retention.
Related methods and services
FAQ
Can AI determine that an ad is fraudulent?
No. Use it to find candidates and organize checks; people verify the original evidence and decide what action is appropriate.
Can public monitoring see private LINE groups?
No. Do not assume access to private conversations or transactions; relevant organizations must use appropriate authority and procedures.
Where does this fit for a bank?
Public impersonation ads and redirection links can provide investigation leads before a payment. They do not replace customer transaction monitoring.
Is JPY 93.74 billion a full-year total?
No. It is the NPA’s provisional SNS investment-fraud loss total for January–August 2026, published on October 2.