Half of Every HSC Grade Is Unenforceable: NSW Ends Take-Home Work Over AI Detection Failure

NSW Just Admitted AI Detectors Can't Be Trusted With a Grade — So It's Removing the Homework Instead

On Monday, NSW Deputy Premier and Education Minister Prue Car directed the state's Education Standards Authority (NESA) to consider banning unsupervised, take-home assessment for the Higher School Certificate — the qualification roughly 95,000 students are due to sit as the Class of 2027. The reasoning, as reported by Tech Times, is blunt: automated AI-detection tools aren't reliable enough to police the roughly half of every HSC grade currently earned on take-home work, and the state would rather remove the unsupervised environment than keep leaning on detection software that has already produced false accusations. NESA's advice could bring interim changes as early as Term 4 this year, with subject-specific carve-outs — art and design-tech projects among them — expected to follow. NSW says it's the first Australian state to make this move; Car has previously backed AI in the classroom, including the 2024 rollout of the government's own NSWEduChat tool.

Why did detection tools fail here?

Because no detector can prove absence — it can only flag suspicion, and NSW decided that suspicion isn't good enough to fail a student on.

What's notable is that this wasn't a policy failure or a discipline failure — it was a measurement failure. NESA already had rules against AI malpractice on paper. The problem NSW is now acting on is that nobody could enforce those rules with any confidence, because detection tools misfire in both directions: missing genuinely AI-written text and wrongly flagging human-written text, with non-native English writers disproportionately caught in the latter. That's the same reliability gap this site has flagged before in AI content governance coverage — a detector's output is a probability estimate dressed up as a verdict, and probability estimates are a bad foundation for a permanent mark on a transcript.

What does this mean for buyers outside NSW schools?

Any organisation using a detector to make a binding decision — pass/fail, published/rejected, paid/unpaid — is running the same experiment NSW just abandoned.

Publishers screening freelance copy, universities grading essays, and content platforms gating payouts on "human-written" scores are all making the same bet NESA made: that a detector's confidence score is trustworthy enough to justify a consequence. NSW's response wasn't to buy a better detector — it concluded no detector on the market clears that bar for a stakes-this-high decision, and changed the assessment method instead. That's a harder, more expensive fix than a software subscription, which is exactly why it's a meaningful signal: a government education authority looked at the tooling and chose architecture over accuracy claims.

Does moving assessment into the classroom actually solve the problem?

It sidesteps detection rather than fixing it, and only works for the portion of work NESA can physically supervise.

Take-home tasks that can't easily move into a supervised room — major projects, long-form research, take-home portfolios — will still need some answer, and Car's own release flags that subject-specific guidance is still coming. In other words, NSW hasn't solved AI-assisted cheating; it has shrunk the surface area where detection was being relied on to solve it. For any buyer still running a detector in production, the practical takeaway is to ask the vendor exactly what NSW's evidence review already asked: what's the true miss rate, what's the false-positive rate on non-native or edited text, and who validated those numbers — the vendor, or an independent evaluator.

Frequently asked questions

Does this ban AI use in schools entirely?

No. It targets unsupervised, take-home assessment specifically — tasks like art and design-technology projects are expected to be exempted, and NSW continues to use its own NSWEduChat tool in classrooms.

When would the change take effect?

Subject to NESA's formal advice, interim changes could apply from Term 4 this year, ahead of the Class of 2027 entering their final HSC year.

Is NSW the first to do this?

Tech Times reports NSW as the first Australian state to respond to AI's effect on high-stakes credentialing by restructuring assessment conditions rather than relying on detection software.

Sources: Tech Times.