In February 2026, a KPMG Australia partner was fined roughly A$10,000 after the firm caught him using an AI tool to generate answers for an internal exam on — of all things — artificial intelligence. The case was not a one-off. KPMG later disclosed that it had identified around 28 similar instances at the firm since July 2025, and it planned to report the problem in its annual results.
The incident is embarrassing for a Big Four firm. But it is more than that. It is a signal that the line between learner and tool has blurred inside training systems. If a global professional services firm cannot stop a senior employee from outsourcing a compliance module to an AI, what chance does a standard online course have?
Completion Is Now a Marketplace
In the same month that KPMG disclosed its AI cheating cases, Mexico's National Autonomous University (UNAM) announced that roughly 58,000 applicants would have to resit their first-ever online entrance exam after suspected AI-assisted cheating undermined the results. In July 2026, Seoul police broke a proxy test-taking ring for the Test of Proficiency in Korean (TOPIK), involving 101 students and university admissions and scholarship fraud. In June, Vietnamese authorities prosecuted a scheme that sold guaranteed foreign-language certificates through proxy test-taking.
These cases are not disconnected. They are evidence that completion — whether of a compliance module, a language test, or an entrance exam — has become a tradable product. On one side of the market, AI agents and generative tools can now answer assessments, write reflections, and speed-run modules. On the other, proxy completion services still sell guaranteed passes, often on pay-after-pass terms. Both models rely on the same weakness: the person enrolled is rarely the person doing the work.
Why Detection Is Falling Behind
The standard response is to add more detection-based proctoring: browser lockdowns, webcam monitoring, keystroke analysis, and AI-detection tools. These approaches try to catch cheating after it has happened. The problem is that the new cheating tools are designed to evade exactly those signals.
A generative AI assistant can answer questions in the learner's own tone, rewrite content, and pause when the system expects human typing. Proxy rings use real people, so they pass most automated checks. And both services operate on demand, with prices, guarantees, and repeat customers. Detection is not a training fraud prevention strategy; it is an arms race that the attacker can run forever.
The Real Cost Is in the Record
For regulated organizations, the issue is not academic. A safety certificate, a food handler credential, or a compliance training record is used to prove that someone was trained. When that record can be bought or automated, the organization is left with a fiction that looks exactly like evidence.
After an incident or audit, investigators will not ask whether a certificate was issued. They will ask whether the named person was actually trained. A completion timestamp, a passed quiz, and a digital badge cannot produce defensible training records on their own. The question becomes whether the organization can show a verified chain from the learner to the credential.
What Verification Actually Means
The alternative is learner identity verification: confirming the identity of the person at the point of enrollment and re-checking it across the course and at the final assessment. It means documenting participation, access history, and any integrity events so the record is defensible. And it means doing this without making the experience punitive for honest learners.
The goal is not to catch more people. It is to make legitimate completion the default path. When a learner's identity is bound to every module, there is no room for a proxy or an AI agent to substitute. The certificate becomes evidence of a real person, not just a real document.
The Path Forward
AI cheating is no longer a curiosity. KPMG's own AI caught an AI being used to cheat on AI training. UNAM's 58,000 retakes show how fast one failure scales. And the proxy rings in Seoul and Vietnam show that the market has been there for years.
The lesson is not that learners have become more dishonest. It is that the systems we built to verify training assumed a world where a human sat at a keyboard and did the work. That world is gone. The organizations that keep issuing credentials on that assumption are not running training programs. They are running record factories.
The next step is to make the record itself defensible. Asgard Authenticate was built to verify the learner, not the login.
Authoritative Sources & Further Reading
- The Guardian — KPMG partner fined over AI exam cheating— KPMG Australia partner fined A$10,000 after using AI to answer an internal AI training exam
- The Guardian — UNAM exam cheating scandal— Mexico's UNAM orders 58,000 applicants to resit online entrance exam after suspected AI cheating
- Chosun Biz — Seoul TOPIK proxy ring— Seoul police uncover proxy test-taking ring for Korean language proficiency test
Writing on training integrity, learner verification and compliance record-keeping for Asgard Authenticate.
Ready to Issue Credentials You Can Defend?
Join organizations worldwide that trust Asgard for online training credential integrity. Start your free trial today—no credit card required.
Start Free Trial