Tokyo Students Experience How a “Dark Job” Recruitment Trap Unfolds

Rei has disappeared

The students do not begin with a lecture. They begin with an absence.

A friend called Rei is missing. Working on their tablets, they search a simulated social-media world for clues. Posts that once looked like background detail acquire a second meaning. A recruiter has noticed something useful: perhaps an urgent need for money, a moment of isolation, a readiness to reply. A private exchange offers attractive, vaguely described work. The conversation migrates away from public view. Personal information changes hands. By the time the true task appears, the recruiter knows enough to threaten Rei and the people around her.

“Rei's Disappearance,” created by Tokyo education startup Classroom Adventure, reverses the usual safety lesson. Instead of giving students a finished list of prohibitions, it asks them to reconstruct the sequence that made the danger possible. Its three instructional stages are translated roughly as: avoid being targeted, avoid being deceived, and avoid becoming trapped. The last stage asks why leaving becomes difficult and how a student can protect a friend as well as themselves.

Bunkyo Ward's official notice says the traveling classes run from July 2026 through March 2027 in ward junior high schools, using tablets to let students experience how social-media posts can connect to crime. The company says the program is going to nine of the ward's ten municipal junior high schools and describes a rollout beginning in June. The one-month discrepancy is small but worth recording: the ward's public schedule is July to March; the vendor describes the broader introduction as starting in June.

The initiative places a young company founded by Keio University students in an unusually serious classroom role. Classroom Adventure was established in October 2024. Its chief executive, Zentaro Imai, and its instructors use the language and interfaces of online life to teach media literacy through games. That proximity to students' digital culture is a strength. It also makes evidence, safeguarding and data practice more—not less—important.

A “job” that is not employment

The Japanese phrase yami baito is often rendered as “dark job,” “shady part-time work” or “black-market gig.” The slang is memorable, but it is misleading. It places the approach beside ordinary work and implies that the problem is a risky corner of the labor market. Japan's National Police Agency now uses the more exact term hanzai jikkōsha boshū jōhō: information recruiting people to carry out crimes.

Police warnings are blunt: this is not a part-time job. Recruits are treated as disposable. They may be sent to collect cash or bank cards from a fraud victim, withdraw money, drive a vehicle, enter a home, watch a location or participate in a robbery while the people directing them remain distant. If one performer is arrested, another can be recruited through the same channel.

The linguistic correction matters in a classroom. “Bad job” suggests a bad choice of employer. Criminal recruitment describes a system that uses the visual grammar of work—pay, task, schedule, application, supervisor—to obtain a person who can absorb the danger of a crime. The apparent wage may never be paid. The “hiring manager” may be an anonymous handler. The application may really be an intelligence-gathering exercise.

That does not erase responsibility for crimes committed. It explains why prevention cannot rely on scolding. A student who thinks only foolish people answer obviously criminal advertisements will not recognize a polished message that begins with delivery, event staffing, document collection or driving. A student who thinks the police will punish them the instant they disclose an ID image may keep obeying the person making the threat.

The trap is a funnel, not a single click

No single row is the whole crime. That is why a reconstructed social-media trail can be more instructive than a poster of keywords. The handler is managing perception. The advertised task stays ambiguous long enough to feel plausible. The request for identity is presented as normal onboarding. The threatening message arrives only after the handler has discovered which threat will work.

The police have published cases showing both routes into the system: people who understood that a task was criminal when they joined, and people who had weak or no awareness until after information was surrendered. In one police example, a minor sent an image of a My Number card, refused the task, was threatened because the home address was known, fled and was later located and protected by police. The point is not that ID makes escape impossible. The case demonstrates the opposite: the threat is serious, and police intervention can break it.

The vocabulary keeps changing

Police examples have included phrases equivalent to “high income,” “¥50,000 a day,” “cash immediately,” “delivery,” “driver,” “document work,” “easy,” “white job,” “available today” and requests for a driving license. Recruiters may exchange public messages on X and then move the conversation to an app such as Signal. Code words can make a task sound childish or harmless.

These indicators are useful, but a memorized blacklist expires. Once a phrase is publicized, recruiters can misspell it, wrap it in an image, borrow legitimate business language, use an influencer account or approach a target directly instead of posting an advertisement. Generative text and mass messaging lower the cost of producing many plausible variations.

The durable skill is verification. Japan's labor ministry says online recruitment information should identify the recruiting entity's name, address and contact, the work, workplace and wage, and must not be false or misleading. A student should search for the company independently, call a number found independently, check whether the address and business exist, compare the pay with the task, and ask why an employer needs an encrypted chat or an ID image before providing a lawful contract.

Even that is not perfect: companies can be impersonated and websites copied. The classroom lesson should therefore teach a stack of tests rather than one magic sign. Ordinary employment tolerates questions, leaves a record, names a responsible organization and does not threaten the applicant for pausing. A handler tries to monopolize time, information and contact.

From cash-collector crews to “Tokuryu”

The present crisis did not begin when teenagers discovered social media. Japanese special fraud developed a division of labor over decades. Telephone callers impersonated relatives, officials or institutions. Other performers collected cash or bank cards, withdrew proceeds, provided accounts, transported people or watched victims. Those visible roles were dangerous to the organization because surveillance and victims could identify them.

Digital recruitment solved a replenishment problem. Instead of relying only on stable membership, organizers could advertise outward, screen replies, separate roles and replace an arrested collector. By 2020, police reports described official X replies warning accounts that posted terms such as “cash collection” and yami baito. The expression had already entered both the criminal marketplace and the counter-message.

A wave of robberies and fraud cases in 2022 and 2023 exposed how remotely directed performers could be assembled through social media. In March 2023, the government adopted an emergency plan specifically addressing robberies and special fraud using social-media recruitment. In September, the Internet Hotline Center began accepting criminal-perpetrator recruitment information for deletion requests.

Police increasingly described the organizational model as tokuryū, an abbreviation for “anonymous and fluid criminal groups.” Unlike a fixed hierarchy whose membership is stable and visible, these networks conceal leaders, connect and recombine, and bring in execution-level people for particular acts. The label is not a new criminal offense and should not be used as a catch-all for every suspicious job. It describes an operating form.

After a further series of metropolitan-area robberies in 2024, the government announced additional emergency measures in December. Police adopted procedures in January 2025 allowing assumed-identity investigations in which an officer can pose as an applicant, under rules and in appropriate cases, including the use of fictitious identification information. The measures were folded into a wider anti-fraud package in April 2025. By 2026, the response had expanded at both ends: covert investigations sought the people recruiting, while middle-school classes sought to reduce the pool they could recruit from.

Young people are visible at the dangerous end

The strongest youth figures must be read with their denominators attached. In 2024, police recorded 21,043 special-fraud cases and approximately ¥71.88 billion in losses. They arrested 416 juveniles in connection with special fraud. Of those juveniles, 286—68.8%—were cash collectors. Juveniles made up 20.7%, about one in five, of all people arrested as collectors.

Those numbers do not say one in five Japanese collectors is recruited by a specific social-media method, nor that one in five young people is at risk. They describe arrested roles in detected special-fraud cases. Arrest data also illuminate the people easiest to catch more clearly than the people directing them.

A separate 2025 police analysis covered people arrested for principal fund-raising crimes thought to be associated with anonymous, fluid criminal groups. The largest single age band was the early twenties, at 23.4%, and roughly six in ten were under 30. About three in ten reported social-media criminal-recruitment information as their route in. Only around one in ten occupied principal or instruction-level positions. This is an organizational portrait: many young performers at the exposed edge, few leaders within reach.

The wider fraud market is expanding for reasons that go beyond youth recruitment. Police recorded 27,832 special-fraud cases and ¥142.31 billion in losses in 2025, with impersonation of police officers a major driver. It would be wrong to attribute that near-doubling in losses to dark-job recruiting alone. But the scale of proceeds helps explain why criminal networks invest in a continuing supply of callers, collectors, accounts and couriers.

Numbers should not turn students into a suspect class. Most young people are not offenders. Some recruits are adults. Some participants knowingly pursue crime; others are deceived and then coerced. Prevention works best when it preserves all of those distinctions while making the immediate risks unmistakable.

Why the detective structure may work

“Rei's Disappearance” uses narrative distance. At first, students are helping someone else. That can reduce the defensiveness produced by a direct question such as, “Would you fall for this?” It also lets a class pause at multiple points and ask what evidence was available then—not what seems obvious after the ending is known.

The reconstruction can expose incremental commitment. Rei does not wake up intending to join a robbery. A post is noticed, an offer is answered, a chat app is installed, a document is sent, a task is reframed, a threat arrives. Each step changes the cost of the next one. In behavioral terms, the handler is narrowing the perceived option set. In moral terms, the student needs to see that an earlier mistake does not transfer control of the future to the handler.

The “protect a friend” perspective is especially valuable. A teenager may dismiss advice about personal vulnerability but notice a friend's sudden secrecy, unexplained cash, fear of a message, travel at odd hours or insistence that no adult can be told. A class can rehearse language that keeps contact open: “I am not going to post this or judge you. Let's save the messages and call someone together.”

Yet immersion has limits. A mystery can make a criminal system feel coherent and solvable, when real approaches are noisy, boring and contradictory. Students may remember the dramatic reveal but not the verification steps. They may also infer that being “targeted” is caused by posting the wrong emotion. The debrief must say plainly: reducing exposed personal data can reduce risk, but public distress never gives a recruiter permission to exploit it.

Promising reactions are not proof of prevention

Classroom Adventure reported that about 450 students at Tokyo Metropolitan Adachi Nitta High School took the program in March 2025, followed by a Metropolitan Police lesson. According to the company's release, 95% said their knowledge had deepened. A Bunkyo public event in December 2025 produced an even stronger immediate survey: the ward reported that every respondent said understanding had deepened, 87% chose the highest rating and 90% enjoyed the event.

Those are encouraging reactions, not evidence that crime was prevented. The Adachi figure was vendor-published and self-reported. The Bunkyo event was a self-selected public session with participants ranging from age six to their seventies; about half were elementary-school children. Neither result establishes what municipal junior-high students remember months later, whether they can detect a novel message, whether they seek help sooner, or whether a live disclosure is handled safely.

The program's growth is nevertheless notable. The company says its broader “Rei's Blog” media-literacy experience has reached more than 50,000 people in at least 14 countries. “Rei's Disappearance” has been used by schools and public bodies in Tokyo, Hyogo and Tottori. In June 2026, Arakawa Ward announced classes at all ten of its municipal junior high schools. Bunkyo's nine-school project and Shibuya high-school classes followed as prevention moved from special events toward repeated public-school deployment.

Scale increases the case for independent evaluation. A practical study could compare baseline recognition with immediate and three-month performance, using unfamiliar ads rather than screenshots from the lesson. It could measure whether students verify an employer, resist moving channels, avoid sending ID, preserve evidence and choose an appropriate source of help. Results should publish denominators, absences and missing responses, and avoid asking children to disclose personal involvement merely to create a dramatic metric.

The most important outcome may be help-seeking efficacy: after a threatening message, does the student believe an adult or police officer can still help without first demanding moral perfection? A prevention lesson that raises recognition but raises shame can leave the most endangered student quieter.

“You can still turn back” is the life-saving line

Public warnings often say that once personal information has been sent, escape becomes difficult. That is accurate as a description of the handler's leverage. It is dangerous if students hear it as fate. Criminal recruiters want the target to believe the door has closed. Police say the opposite: even if a person or family is threatened, do not participate; consult police.

The exit should be practiced in class as concretely as the recruitment. Students can rehearse closing the chat, taking screenshots without spreading them, moving to a teacher, saying the first sentence to #9110 and helping a friend stay physically accompanied. They should know the difference between 110 for urgency and #9110 for consultation.

Schools need their own protocol before the first tablet is opened. A student may disclose that an approach is current, that an older sibling is involved or that a family address has been exposed. The teacher should not turn the disclosure into a public teachable moment, seize control through punishment, or promise absolute confidentiality that cannot be kept. The response needs a named safeguarding lead, private assessment of immediate danger, evidence preservation, police and guardian coordination as appropriate, and a plan for travel and home safety.

Content warnings and an alternative activity matter too. A disappearance narrative, threats to family and robbery material can affect students who have experienced victimization or family involvement. An opt-out should not require public explanation. A counselor or private follow-up route should be visible after the class, when a student may finally decide to speak.

The platform battlefield is measured in posts, not victories

Education operates downstream. The recruitment post, direct-message system and anonymous chat operate upstream. In 2024, the Internet Hotline Center identified 13,852 items of criminal-perpetrator recruitment information. Excluding material already gone, it requested removal of 9,234 items, and 7,831 were removed. Those are pieces of content, not unique recruiters or prevented crimes; one network can create many posts and return under new accounts.

Police have also used automated systems to find suspicious posts and reply with warnings. From April through December 2024, a National Police Agency system sent 4,807 warning replies. In 2025, police reported 6,679 deletion requests for criminal recruitment information and 6,351 completed removals, roughly 95%. Faster removal reduces exposure, but a removal rate does not measure how long the post was live, how many people saw it or where the conversation moved.

Platforms can reduce the burden placed on a 13-year-old. They can verify high-risk recruiting advertisers, disrupt repeated account creation, detect image-based and euphemistic offers, add friction before an account sends many unsolicited messages, preserve evidence for lawful investigation, and place help prompts inside risky channel-migration patterns. They must also provide appeals so automated enforcement does not erase legitimate small employers or researchers.

Lawful job information has a governance structure. The labor ministry's required identifying information gives platforms a starting point for advertiser checks. Article 63 of the Employment Security Act criminalizes recruitment for illegal acts, including coercive use of personal or family information in this context. The law is not merely telling young people to be more skeptical. It assigns responsibility to the people who recruit them.

A device in every hand creates a data question

The product page says the simulation runs in a browser and generally requires one web-connected phone, school computer or tablet per student. That makes deployment easy. It also means a school should know what the browser sends before asking minors to participate.

Classroom Adventure publishes a general privacy policy and an information-security policy. The privacy page describes ordinary website data such as IP address, browser, provider, time, pages and clicks, and says it does not knowingly collect personal information from children under 13. The security page describes organization-wide principles. Public material reviewed for this article does not provide a game-specific student data sheet stating whether pupils use accounts or identifiers, which event data are logged, whether answers or chat choices leave the device, exact retention, storage locations, subprocessors, access by teachers or the company, or deletion procedures.

That absence from a public page does not prove the system collects sensitive student data. It means procurement should obtain a precise map. Some first-year junior-high students may still be 12, which makes the under-13 statement operationally relevant. The safest design is data minimization: no account where none is needed, pseudonymous classroom sessions, no unnecessary free-text disclosure, short log retention, school control of rosters, no advertising use and a clear incident route.

There is also a pedagogical privacy issue. A lesson about money trouble should not ask students to reveal their own debts, family conflict or prior messages in front of peers. Scenario performance can be evaluated without creating a new database of children's vulnerabilities.

Prevention cannot end with individual vigilance

The first step of the simulation—avoid being targeted—can easily slide into advice to hide financial distress. Privacy helps, but secrecy about hardship can also isolate the person a school most needs to reach. The recruiter is exploiting low income, debt, unstable work, loneliness, conflict or a desire to belong. Telling a teenager to post less does not remove those conditions.

A complete prevention strategy pairs digital literacy with accessible legitimate work and support. Students need to know where to ask whether a job is real, where to discuss debt or family stress without punishment, and how to reach youth services privately. Employers and job platforms need fast, understandable pathways for first-time workers to verify offers. Families need language that invites disclosure before demanding an explanation.

Schools should not teach that every high-paying flexible job is criminal or that encrypted technology is inherently illicit. Journalists, activists, businesses and families use privacy tools lawfully. The red flag is the pattern: unverifiable identity, implausible reward, vague work, secrecy, forced migration, premature ID collection, urgency, task substitution and threats.

The best question is not “Do I recognize this dark-job keyword?” It is “Can I independently establish who is asking me to do exactly what, where, for what lawful organization and under what terms—and can I pause without being threatened?”

What Bunkyo should measure before March

Bunkyo has an opportunity because the program runs across an academic year rather than a single auditorium event. It can refine teacher training, observe which clues are misunderstood, test retention and build a disclosure route that remains after instructors leave. It can also publish what did not work.

That transparency would benefit Arakawa, Shibuya and other municipalities considering similar experiences. It would also benefit Classroom Adventure. A well-evaluated Japanese serious game could become more than an engaging local lesson; it could show which parts of a coercion sequence young people can actually interrupt.

The lesson after the mystery is solved

Rei's story ends when the students assemble the clues. A real dark-job approach does not end with recognition. It ends safely only when the target has somewhere else to go.

That is the deeper promise of the Bunkyo experiment. The students are not being trained to become tiny detectives who must outwit organized crime alone. They are learning to notice when an ordinary-looking work conversation begins stripping away verification, privacy, time and choice. The institution around them must then prove that asking for help restores those things.

Japan's history shows why the challenge persists. Fraud groups divided dangerous tasks, recruited replacement performers, hid leaders behind remote channels and adapted their language faster than warning posters could be printed. Police and platforms responded with deletion, warning replies and undercover investigation. Schools are now trying to reach potential recruits before the first message.

The classroom can make the sequence visible. It can give a student the words to interrupt it. It can let a friend practice staying beside someone who is frightened. But it must not reproduce the recruiter's greatest lie: that a surrendered document or one wrong reply makes the next crime inevitable.

The mystery's final answer should therefore be a route, not a slogan: stop, save the evidence, tell someone, call for help, stay together. “Dark job” recruitment works by making a person feel alone and already lost. The most powerful simulation is one that rehearses the moment a door opens.