Artificial intelligence has become the hottest technology sector, and related training programs are springing up everywhere, from established education firms to various other companies. In March of this year, Feng Yufan, who was job hunting in Shenzhen at the time, received an unexpected phone call.

The caller claimed to be an HR representative from Shenzhen XXXX Talent Service Technology Co., Ltd. (referred to as Company X), a firm specializing in direct recruitment for medium and large enterprises, offering a position in the AI field. However, because Feng lacked relevant project experience, he would be required to complete about three months of "free pre-employment training." Feng's major and work history had nothing to do with AI, and he had never applied for such roles, yet the HR person on the phone assured him that even with zero background, they would "definitely teach you until you're ready, guaranteeing 100% job placement."

Feng recorded the conversation. In it, the caller stressed that Company X "only provides employment services, not training," while simultaneously promising to offer Feng courses on AI large models. A company that claims not to be a training institution but offers training courses — the contradiction puzzled Feng. In fact, Company X doesn't appear on the list of registered training institutions published under the "Training Institutions" section of the "Shenzhen Human Resources" mini-program, the official platform of the Shenzhen Bureau of Human Resources and Social Security. Toward the end of the call, the HR person told Feng that if he completed the large model course and successfully found a job, he would owe the company an employment service fee: "After you get a satisfactory offer, you pay the first part, which is 16,900 yuan; the second part comes after you actually start working and receive your salary, paid in installments of 2,450 yuan per month for 8 months."

This wasn't the first time Feng had received such a call. In December 2025, he got a similar pitch from someone using the same "recruitment-to-training" model, promising that a month of training would lead to a job in electronic information. Lured in, Feng attended a month of classes in an office building in Bao'an District and took out a loan to cover the tuition. Realizing something was wrong just before the Spring Festival, he filed a complaint with the relevant authorities and cancelled the loan agreement. That company has since ceased operations. Multiple regions have begun cracking down on these practices. In May of this year, five government bodies — the Ministry of Human Resources and Social Security, the Office of the Central Cyberspace Affairs Commission, the Ministry of Education, the Ministry of Public Security, and the National Financial Regulatory Administration — issued a joint risk warning, urging job seekers to be wary of "recruitment-to-training" schemes and "training loans."

Section 1: "Guaranteed Employment"?

Having already been deceived once, Feng was skeptical of Company X's promises of "monthly salaries above 10,000 yuan" and "guaranteed jobs." Fresh graduate Zheng Ming wasn't so fortunate. Earlier this year, Zheng paid over 20,000 yuan in "service fees" to Company X, endured three months of so-called "AI large model training," and then, with the company's guidance, fabricated his work history and project experience to land an AI application job — only to be fired two months later.

Even at well-known training institutions, the situation for trainees is often bleak. In July 2025, software engineer You Wei, 35, enrolled in a large model course at Institution A in Shenzhen. Institution A is a veteran training provider, famous during the internet boom for its programming courses. You Wei paid 17,000 yuan in tuition, which included job placement services. Institution A had requirements for placement-track students: a bachelor's degree and two years of relevant work experience. You Wei qualified, so he resigned from his job to attend. After the course ended, he followed his placement counselor's instructions and falsified his resume, which got him through an interview, but the background check later exposed the fabrication, leaving him deeply shaken. He's since gone back to work as a software engineer.

But a quick scroll through social media tells a very different story. In countless posts posing as "personal experiences," AI short-term training appears to lead to a rosy employment landscape: someone claims to have found a well-paying job earning over 10,000 yuan a month after just a month of AIGC training, complete with photos of pay stubs and deposit alerts. Another claims that after three months of AI large model training, they landed a job at a major logistics company with a five-figure monthly salary.

Finding a real trainee to talk to before writing this article proved surprisingly difficult. When I reached out to these posters, posing as a job seeker looking to learn AI, their claims grew increasingly dubious: "Found a job less than a month after finishing," "Even liberal arts majors can learn it," "High school graduates can do it," "One person got a job paying 50,000 a month." Given today's job market, their stories sounded almost too smooth to be true. After enough conversations, patterns emerged: accounts with anime avatars tended to be promoting training at Company M, while those with attractive female profile pictures were usually pushing courses from Company N. Feng Yufan told me he'd posted a warning about the "recruitment-to-training" trap online, but the post was quickly reported and taken down. He later managed to reach several former Company X trainees, some of whom had shared their real experiences on social media — all of which were also reported and deleted.

Section 2: The Overnight Success Myth in the AI Boom

Zheng Ming's path to Company X started when he contacted another institution about courses after seeing their social media ads, and then mysteriously received a call from Company X. The pitch echoed what Feng had heard. At the time, Zheng was back in his hometown, anxious about unemployment. His engineering major had narrow career prospects and low pay, and he saw AI as his ticket to break into the internet industry. The Company X staffer promised job placement after training, and Zheng didn't press for details — he was so desperate for a career change that he packed his bags and headed to Shenzhen.

At Company X's promotional seminar, several "former students" told Zheng they'd all found jobs and showed him their offer letters. Looking back, Zheng isn't sure whether those "alumni" were hired actors. Five days into his time at Company X, after a trial session, he signed an "Artificial Intelligence Employment Consulting Services Agreement." The contract stipulated a total service fee of 36,000 yuan: roughly 45% due upon receiving an offer, with the balance spread over eight months of installments. Zheng then began a three-month training program. His class had over forty students from varied academic backgrounds, many of them liberal arts majors. During his three months there, Company X ramped up its class openings — initially one new class per month, then one every two weeks, and by the end, one every week.

Company X offered both "AI large model" and "AI application development" tracks, but Zheng noticed the curriculum was identical for both. In his view, the training essentially taught students how to use AI large models for coding. The first month covered basic programming concepts. Zheng had studied some coding in college and found the material shallow and outdated, completely out of sync with what the internet industry actually needs. The second month moved to project practice, but the instructor merely walked through a project's architecture — students rarely got hands-on opportunities. The class didn't provide access to large models either; students had to register for foreign accounts through their own channels, and even domestic models required them to buy tokens out of pocket. After finishing the course, Zheng taught himself large model concepts using free B站 resources and open-source projects. Comparing the two, he felt Company X's training was like a history teacher reading only the table of contents — superficial, with no real engineering mindset instilled.

You Wei similarly felt his training at Institution A was lacking. The curriculum covered text classification, RAG, Agent, and Langgraph — all part of the AI application development stack — but "some of those techniques are rarely even asked about in interviews anymore," he said. Online, others have also questioned the real value of such training. Companies in China capable of large model development are either the big internet giants or dedicated AI research firms like DeepSeek, Moonshot AI, and Zhipu AI. Their core R&D roles have extremely high hiring bars — strict requirements on academic background, degree, and university prestige — and no training institution can help trainees clear those hurdles. For those simply wanting to learn AI application development, the internet is full of high-quality free or low-cost resources that far outperform what training centers offer.

Hada, an AI product director at an internet company, put it bluntly in related discussions: "If I were hiring an AI product manager and saw a training-institution certificate on their resume, that would be a red flag for me." He believes people who've genuinely accumulated experience in AI applications or products don't learn from training courses. "I'd prefer someone who's self-taught — someone who digs through source code, finds collaborators, or figures things out on their own." Hada and his colleagues typically explore AI applications through overseas communities, company blogs, academic papers, open-source code, and trying out new products. When evaluating AI product managers, he looks beyond AI coding skills to whether candidates have the mindset and practical experience of using AI to solve problems. Simply put, "that person can use AI tools — not just AI coding tools — to independently complete work that would normally require several people."

Section 3: Young People Who Don't Survive the Probation Period

By May, former trainees started trickling back into Company X's office. When Zheng asked around, he learned they'd all been laid off by the companies that hired them. It dawned on him that something was wrong. For those who were laid off, Company X provided a consultation room where they could revise their resumes and apply for jobs. Trainees could use the company's resume templates to fabricate project and work experience, then target small companies — "because small firms usually won't pay for background checks, so everyone just gambles on that." The Company X training experience itself was never mentioned on any resume. After finishing the three-month course, Zheng joined his classmates in "polishing" resumes and memorizing interview scripts. He landed a job in under a month but was laid off after just over two months.

He considered continuing to self-study, but AI technology evolves so quickly that he couldn't keep pace, and he felt he simply couldn't break into the field. Of the trainees Zheng knew, most were let go within about a month of starting their jobs; only a handful scraped through probation. But even those few didn't credit Company X's training. Some had a background in development and worked incredibly hard, "studying on their own until two or three in the morning," while others who changed careers barely managed to stay by self-teaching and building good relationships with colleagues. Among Zheng's classmates, some paid the full fee upfront, while others, like him, paid over 10,000 yuan for "employment services" upfront and agreed to installment payments for the rest. Some never got offers at all, yet Company X claimed "the trainees weren't trying hard enough, not that the teaching was at fault," and demanded payment. Others who were laid off during probation and refused to continue paying received legal letters from the company.

After being laid off, Zheng tried again to find AI development roles but could barely secure interviews. He noticed that during the first half of the year, the AI boom had companies in a frenzy — the market was overheated and job demand inflated. But as the market cooled in the second half, hiring demand dropped accordingly. Through one of Company X's trainees, I got in touch with three individuals who took Company X's big data courses in 2025. Their stories closely mirror Zheng's: they completed one-to-fifty-day big data courses, then used Company X's resume templates to fabricate work and project histories. Trainee Hong Wei landed two big data outsourcing jobs in succession but failed probation both times. Trainee Li Chen never found a position meeting the salary promised in the agreement but still paid Company X over 10,000 yuan. Trainee Xiao Lin was asked to leave just days after starting and paid several thousand yuan in service fees. After his training, Hong Wei realized that everything Company X taught could be found for free or at low cost online — the company had simply repackaged scattered resources into a structured format. And it wasn't until he actually started working in big data roles that he understood how inadequate that knowledge was for real-world demands.

Section 4: A Carefully Designed Operating Model

The paths that led Hong, Li, Xiao, and Feng to Company X were identical: while applying for jobs on recruitment platforms, they each received calls from Company X. On the phone, the caller offered big data positions but said the candidates needed to come in for a course. All three confirmed they'd never applied for any big data roles or any positions at Company X before the calls — and none of them knows how their personal information was leaked. Feng Yufan, however, was determined to get to the bottom of it. After that call, he started investigating why he'd been contacted. Reviewing his job applications, he found he'd once applied for a role at Company Y, a position related to his own major but completely unrelated to AI large models. A quick search revealed that all of Company Y's posted roles were unrelated to AI. Digging deeper, Feng discovered that Company Y and Company X share the same legal representative, with registered addresses in Luohu and Longgang districts respectively. When he visited Company Y's registered address, he found no actual office there. He later attended Company X's promotional seminar and saw much of what Zheng had described. Before publication, Company Y's job listings had been removed from the recruitment platform where Feng had applied.

The official Bao'an district platform, "Binhai Bao'an," issued a warning in June 2026: those perpetrating "recruitment-to-training" and "training loan" fraud often operate through two or more seemingly unconnected companies. One handles recruitment and interview funneling, sometimes even impersonating central or state-owned enterprises with fake listings. The other signs the training agreement and induces applicants to apply for loans. This separation of operations is designed to sever the chain of liability, increase confusion, complicate consumer rights enforcement, and evade legal consequences. Over the past few months, Feng has connected with a group of former Company X trainees on social media and, armed with their collective experiences, has filed complaints with the human resources, development and reform, and petitionary departments in both Luohu and Longgang districts. The Luohu District Human Resources Bureau responded that since the actual training and follow-up services took place at Company X in Longgang, and Company Y never actually conducted training, "if there were genuine training activities, we would naturally intervene, but in this situation it's very difficult for us to investigate," and referred the complaint to Longgang. The Longgang District Human Resources Bureau replied that Company Y fell outside its jurisdiction, and that Company X's actions were hard to define as false recruitment or advertising, and equally difficult to classify as violations of labor law. The Longgang District Development and Reform Bureau said its investigation found Company X was not involved in any illegal training loan activities, had no third-party lending partners, and no other financial crimes were detected. This operational model effectively sidesteps substantive legal and regulatory constraints.

Cao Jingshan, secretary-general of the Criminal Law Committee at Beijing Strategy (Guangzhou) Law Firm, has been closely following the legal dimensions of "recruitment-to-training" schemes. She notes that this "separation of recruitment and training entities" creates a regulatory blind spot. Current laws and regulations primarily target entities that combine recruitment and training in one, with no clear rules governing jurisdiction when the two are split. Job seekers file complaints in Luohu, only to be told Company X operates in Longgang; they file in Longgang, only to be told the recruitment funnel goes through Company Y, registered in Luohu. Complaints bounce back and forth until they die, unresolved. Cao emphasizes that even if a complaint reaches regulators, it's hard to handle: complainants often lack a complete evidence chain, making it tough for authorities to make accurate judgements based on partial information. Besides, "recruitment-to-training" is a recent form of fraud with many guises, and agencies need time to identify its illegal nature. Furthermore, the scheme spans multiple areas: false job postings touch on human resources market management; training fees involve education and training market oversight; false advertising falls under market supervision; determining whether it constitutes fraud requires police investigation. No single enforcement body can tackle it alone, and multi-department coordination depends on institutional coordination mechanisms that are currently still lacking.

On the "training loan" classification, Cao analyzes that in 2023, five government departments including the Ministry of Education explicitly banned the use of training loans for fee collection at off-campus training institutions. By labeling its fee a "service fee," Company X avoids scrutiny of training fees. Under regulations like the "Personal Loan Management Measures" and the National Financial Regulatory Administration's risk warnings on false advertising of online loans, a loan involves elements like capital lending, interest agreements, and third-party lenders. However, the agreement between Company X and trainees is an "interest-free deferred service fee" — the trainee receives the service first and pays later in installments, which is a payment arrangement within a service contract. It involves no capital lending, no third-party lender, and in principle does not constitute a loan. By not bringing in a third-party lending institution, Company X avoids classification as a training loan. With no third-party guarantees or insurance products, it also evades guarantee-related regulations. In Cao's view, Company X's fee structure was deliberately designed to bypass both the Ministry of Education's prohibition and financial regulators' oversight of loan products.

Section 5: The Difficulty of Seeking Redress

Zheng Ming never filed a complaint against Company X. He's too anxious about his job hunt to spend more energy, and he worries that a dispute could hurt his future employment prospects. He'd heard that other trainees who tried to seek recourse ran into walls. Even if Zheng did try, he'd face serious obstacles. His signed agreement with Company X is heavily lopsided when it comes to rights and obligations: the company's duty to recommend employment is barely constrained by the contract's terms, while the trainee's payment obligations are firmly locked in.

Article 3, Clause 2 states: "Party A (Company X) promises that after Party B (trainee) completes business training and passes assessment, Party A will recommend employment opportunities to Party B, with positions in artificial intelligence and related fields... After Party B is placed through Party A's employment services (in Shenzhen, Shanghai, Beijing, Guangzhou), the comprehensive pre-probation salary (including subsidies) shall be no less than 10,000 yuan per month for large model application roles and no less than 15,000 yuan for algorithm roles. Party B shall continue to enjoy one year of free technical support follow-up services and re-employment recommendation after employment." The wording here is "recommend employment opportunities" — Party A's obligation is limited to "recommendation." Whether the candidate is hired, passes probation, or earns the promised salary is not listed as Party A's responsibility. Article 5, Clause 1 states: "All employment in this agreement refers to Party B's first receipt of a job offer (including verbal promises or email notifications), or Party B's internship or probation at an employer, or any other form of de facto employment relationship formed with an employer. This includes employment facilitated by Party A's recommendations, as well as employment Party B secures independently by contacting employers during or after the training period." Under this clause, any form of offer — even a verbal one, even just an internship or trial period — counts as "employed." Even more significant: if trainees find jobs on their own during or after training, it's automatically credited as Company X's "employment service achievement," triggering payment obligations.

Article 5, Clause 2 reads: "If Party B's actual employment salary does not meet the standards in the enterprise commitment letter, Party A is obligated to continue providing employment recommendation guidance until the standard is met. If Party B refuses to accept an offer letter, fails to start work for personal reasons, or voluntarily accepts a position with a salary below Party A's promised standard, Party B shall be deemed to have successfully obtained employment, and Party A shall not be in breach of contract." In other words, even if a trainee goes months without finding a position at the promised salary, Company X's only obligation is to "continue recommending and guiding." If a trainee, squeezed financially, accepts a lower-paying job, it's deemed "voluntary acceptance," releasing Company X from liability and triggering the payment obligation. Article 8, Clause 2 states: "Provided Party B cooperates with independent interviews and recommended employment, if Party A has achieved successful employment for Party B — determined by the first offer received and start of work — Party B's refusal to pay service fees after employment constitutes breach of contract." "Determined by the first offer and start of work" means that regardless of whether the trainee passes probation or gets fired, once that first day of work happens, the payment obligation is locked in.

Throughout this contract, Company X's obligations are process-oriented — recommend, guide, follow up — with no guarantee of outcomes. The trainee's obligation, by contrast, is absolute: pay. Any form of employment, even one the trainee finds independently, counts as triggering the payment requirement. Hong Wei and Li Chen, who took Company X's big data courses in 2025, provided copies of their agreements. Except for the tuition amount, job role, and payment terms, the details are nearly identical to Zheng's. Cao Jingshan analyzes that the deliberately vague language in the contract regarding employment promises suggests Company X knows it cannot deliver. By separating its marketing claims from the contract, the company can later argue it never made any "guaranteed employment" promise, or dismiss it as mere "sales talk."

Cao points out that in judicial practice, for a court to recognize "guaranteed employment" as part of the contract, at least one of the following conditions typically must be met: the contract explicitly states specific promises like "100% employment guarantee" or "minimum monthly salary"; promotional materials are appended to the contract; or a service commitment list is included in recruitment materials. But this doesn't mean trainees have no recourse. Even with vague contract language, trainees who can provide a complete evidence chain — screenshots of promotions, recruitment brochures, recorded sales calls, recruitment videos — still have a chance to prove "guaranteed employment" was an implied term of the contract. Increasingly, courts are moving toward reviewing the entire "promotion-signing-performance" chain rather than relying solely on contract clauses. In recent "recruitment-to-training" cases heard in Shenzhen courts, multiple rulings have found "guaranteed employment" advertising to constitute fraud.

Among the trainees I spoke with, none except Hong Wei had fully paid off their "service fees" to Company X. They don't want to pay the remaining balances, but they're also afraid of legal trouble. Cao explains that stopping payments creates two problems. On default risk: once a trainee signs an installment agreement and stops paying, they're in breach. Company X can demand continued payment, seek the full remaining balance immediately (if the contract has an acceleration clause), claim damages, and even pursue litigation costs and attorney fees if specified in the contract. On the difficulty of seeking relief: if trainees want to rescind the contract on grounds of fraud, they must file a countersuit or separate lawsuit to exercise their right of rescission and bear the corresponding burden of proof. If they can't produce sufficient evidence, they may still lose. In Cao's view, even trainees who decide to pursue legal action face multiple practical hurdles. First, under the principle of privity of contract, trainees can only sue Company X, the party they signed with — it's very hard to add Company Y, the recruiting funnel, as a co-defendant, so courts can't easily examine the entire "recruitment-to-training" scheme in a single case. Second, the employment guarantee made orally by Company X's staff never made it into the contract, so once in court, trainees will struggle to prove Company X made any specific "guaranteed employment" promise. Third, Company X did provide training and did help some trainees land offers — even if only a few, and even if those trainees were quickly fired — which is enough for the company to argue in court that it "fulfilled its contractual obligations." At least one court ruling has taken this view.

From the judiciary's perspective, Cao sees three layers of difficulty in such cases. On evidence: trainees struggle to meet the burden of proof, and under the "who claims, must prove" rule, their cases are dismissed. On characterization: courts generally focus on whether "training obligations were fulfilled," making it hard to conduct substantive review of the entire "recruitment-to-training" structure, while the standard for proving "fraud" is extremely strict — plaintiffs must prove all four elements of fraud. On legal application: different legal bases require different evidentiary emphasis, making it hard for trainees to choose the right approach, and courts can only rule based on the plaintiff's stated claims — they can't choose the legal basis on the trainee's behalf. Cao also points out that the "Minutes of the National Court Civil and Commercial Trial Work Conference" calls for "look-through" judicial thinking to uncover the parties' true intentions and the actual legal relationships in civil and commercial cases. The "recruitment-to-training" model should be examined using this approach, but in practice, many obstacles remain.