Quick Answer: The Australian STEM Videogame Challenge, run by ACER, is shutting down after 12 years because judges could no longer reliably tell which student games were made by kids and which were generated with AI. The organiser said generative AI made it harder to judge student work and pulled the program away from its original purpose. The lesson for any student game development competition is that judging the output alone no longer works.
Twelve years of kids building games together didn't end with a scandal. It ended with a judging problem. When the entries got harder to trust, the people running the contest decided the honest option was to close it. If you run a school program, a game jam, or a club, that story should worry you, and it should also tell you exactly what to change.
What Happened to the Australian STEM Videogame Challenge
According to a report in PC Gamer, which Kotaku summarised, the Australian STEM Videogame Challenge will not run again after this year. The competition was open to students from elementary through high school. Students worked alone or in groups of up to four, and submitted a game built around a chosen theme, plus a design document and a short trailer.
That format is sensible. A design document shows intent, and a trailer shows the game running. Together they were meant to prove that students had built something real. The problem was that a polished trailer and a tidy design doc are exactly what current generative tools can produce in quantity. The format measured the artefacts, and the artefacts stopped being a reliable signal.
Cathy Zemaitis, head of the Student Learning and Progress division at ACER (the Australian Council for Educational Research), which runs the challenge, put it plainly in her statement to PC Gamer: the decision was driven by "the rapid and complex rise of generative AI in game development, making it harder to judge what students made themselves, and moving the program away from its original purpose."
Notice what she didn't say. She didn't say AI was banned or that students were cheating at scale. She said the judges couldn't tell, and that the program had drifted from its goal. That distinction matters, and we'll come back to it.
Why Judging AI Use Is Harder Than It Looks
Here's the core problem. In a professional studio, using AI to make games faster and cheaper is a business decision that's openly discussed. In a classroom, the goal is different. The purpose is developing skills that haven't formed yet. A 12-year-old who asks a model to write a state machine has skipped the learning, even if the game runs.
Judges face a verification problem. Looking at a finished game, a trailer, and a design doc, you can't reliably see the path that produced them. Was the sprite drawn by hand or generated and touched up? Did the student write the collision code or paste it in and debug it? Those questions have no answer in the final files.
Most people assume AI detection tools will solve this. They won't, at least not in a way you can defend to a parent. Text and image classifiers produce false positives and can be evaded with light editing. Using one as the sole basis for disqualifying a child is the kind of decision that ends up in a complaint letter. Treat detectors as a prompt for a conversation, not as evidence.
The Counter-Intuitive Part: Banning AI Is the Lazy Fix
The popular response to stories like this is "ban AI in student work." It sounds decisive, and it's incomplete for three reasons.
First, a ban is only as strong as your ability to enforce it, and as the ACER situation shows, output-based enforcement is weak. Second, students will encounter these tools everywhere, including in the professional tools they'll be expected to know. Third, a blanket ban pushes usage underground, where you learn nothing about how students worked.
The better move is to change what you measure. Judges who only see the finished game can't distinguish a student who built it from one who prompted it. Judges who see the development process can. That means version history, dated commits, build logs, a short dev diary, and a live walkthrough. None of these are foolproof, but together they make the work visible in a way a trailer never will.
This next part matters more than it looks: process evidence also teaches. A student who has to explain why they chose a particular physics approach is practising the exact reasoning the competition was meant to build.
What the Parent's Account Reveals About the Real Cost
The most affecting part of the coverage isn't the statement from ACER. It's the Reddit post from a parent whose teenage son had entered the competition for several years. The parent described their son and friends "meeting up regularly, discussing ideas, and working together on a shared passion project," and said it gave them "a common goal to strive towards outside of school."
Read that again. The value wasn't only the finished games. It was the group, the recurring deadline, and the sense of belonging. Those are things a judging rubric doesn't score, and they're exactly what gets lost when a competition closes.
The parent also wrote, "My son was visibly gutted when I told him the news, and it genuinely felt like part of his identity had been destroyed." That's a strong emotional statement, and it's one person's experience, not a survey result. Still, it points to a practical lesson: when you change or end a youth program, the people inside it feel that loss first. If you're running something similar, communicate early, explain the reasoning, and say what's coming next.
Kotaku's coverage also framed the story around a broader worry about the industry. The writer argued that adults normalising generative AI risks eating the "seed corn" of game development talent. That's an opinion, and a reasonable one to debate, but it isn't a measured finding. Keep that distinction in mind when you read commentary on the topic.
How Competitions Can Judge Process Instead of Output
If you're designing or running a student game competition now, you need a judging model that doesn't depend on guessing. The table below compares the common approaches.
| Judging approach | What it verifies | Main weakness | Best use |
|---|---|---|---|
| Final submission only (game + trailer + design doc) | Polish and finished presentation | Can't distinguish student work from generated work | Showcase or audience vote, not skill assessment |
| AI-content detection tools | Statistical patterns in files | False positives and easy evasion with light edits | Flagging for a follow-up conversation, never sole evidence |
| Version control and dev logs (e.g. Git commit history, dated build logs) | Incremental work over time | Students can backfill or commit in bulk | Core evidence for team and individual contributions |
| Live walkthrough or short oral defence | Whether the student can explain design and code choices | Time-intensive for judges with many entries | Finalists or disputed entries |
No single row is enough. The strongest setup combines dated process records with a short live defence for shortlisted entries. That raises the workload for judges, which is a real cost, and it's why many small programs struggle to adopt it. Plan for that trade-off instead of pretending it's free.
Here's a practical starting point: require a dated development log from the first week, ask for at least one commit or build per week, and reserve live defences for the top few entries. Explain the rationale in your rules, so students understand that you're assessing learning, not policing them.
Where This Fits in the Wider Games Industry
This competition isn't an isolated case. Kotaku's related coverage notes that itch.io has deleted what it described as "terabytes" of AI slop as it tried to counter a surge in game uploads, and that its AI disclosure policies are part of that effort. Kotaku also reported that 2026 had already seen more titles added to Steam than all of 2025, with around 24,000 titles expected by year's end.
Those numbers explain the pressure. When storefronts are flooded, platforms reach for disclosure rules and takedowns. Competitions face a smaller version of the same problem: volume rises, and manual judging gets harder. Disclosure is a reasonable policy, but it depends on honesty, which is why process evidence matters more than a checkbox.
The fair counterpoint is that some developers use generative tools openly and legitimately, including for prototyping and asset cleanup. A blanket view that all AI use equals slop is as lazy as a view that none of it is. Your competition rules should say which uses are allowed and why, and they should match the goals of the program.
Frequently Asked Questions
What is the Australian STEM Videogame Challenge?
The Australian STEM Videogame Challenge is a competition for young people run by ACER, the Australian Council for Educational Research. Students entered games built around a chosen theme, with a design document and trailer. It ran for 12 years before organisers announced it would close after this year, citing difficulty judging AI-generated entries.
How do you tell if a student game was made with AI?
You can't reliably tell from the final files alone. Detection tools give false positives and are easy to evade. Better evidence comes from dated version history, development logs, and a short live walkthrough where the student explains design and code decisions. Use detection as a reason to ask questions, not as proof.
Why did the student game competition shut down?
ACER said generative AI made it harder to determine what students made themselves, and that this moved the program away from its original purpose of building real-world skills. Judges were no longer confident they were assessing student work, so the organisers chose to end the program rather than keep judging unreliable entries.
Can students use AI tools in game development competitions?
It depends on the rules, and the rules should reflect the competition's goals. A skills-focused contest may ban generated code or art outright, while a more open event might allow tools with disclosure. Whatever you choose, state the rationale clearly and require process evidence so you can check compliance.
What is the best way to judge student game development AI use fairly?
Combine several signals instead of relying on one. Require dated commits or build logs, ask for a short dev diary, and hold a live defence for shortlisted entries. This approach makes the work visible over time, which is far harder to fake than a finished trailer, and it doubles as a learning exercise for students.
Running or entering a student program right now? Pick one change this week: add a dated development log requirement to your rules and see how many students can produce it without scrambling. Share this with a teacher or club leader who's wrestling with the same judging problem, and ask them what evidence they'd trust most.