What the £875,000 Cambridge English Fine Reveals About Language Assessment Technology

On 11 June 2026, Ofqual fined Cambridge English £875,000. The story was framed everywhere as a technology failure. The real lesson is somewhere else.

How 93,865 Scoring Errors Went Undetected for Two Years

Between August 2023 and September 2025, automated marking errors in the listening and reading sections of IELTS produced incorrect results. The system marked 93,865 responses wrong. 62,794 candidates received corrected results, and most had originally been scored too low.

No routine check caught the problem. It surfaced during an update to the monitoring system, almost by accident.

These are not abstract numbers. IELTS scores decide university admissions, job offers, and visa applications. Half a band too low can change where someone is allowed to study, work, or live.

The Real Cause: Automation Without Oversight

It is easy to blame the automation. That misses the point.

IELTS marks listening and reading by comparing each answer against a fixed key set by human experts. For closed questions with one correct answer, that is the right method. Not an outdated one.

The failure was not the choice to automate. It was that no one verified the results for two years. Ofqual was direct: the weakness was inadequate monitoring and error detection.

Every system produces errors. The real failure is not that one happened. It is that it took two years to find it.

Automating Language Tests Is Not the Problem. Skipping Checks Is.

The obvious reaction is to put humans back in charge of every score. That does not hold up.

Manual marking does not scale to millions of tests. It is slower, more expensive, and less consistent from one marker to the next. It also keeps reliable assessment out of reach for most candidates, who cannot wait weeks or pay for a premium result.

Automation is not optional anymore. It is the only way to make assessment fast, consistent, and accessible.

But automation without oversight is exactly what failed here. So the role of people changes. They move from producing scores to validating them: reviewing results, running trials, checking outputs against expert judgment, continuously, so the system keeps improving instead of drifting in silence.

A test is a measuring instrument. Like any instrument, it drifts. Recalibrate it, or you end up trusting a number you have no reason to trust.

How Talketet Builds Accountability Into Every Score

At Talketet, quality control is built into the product, not added afterwards. The platform assesses speaking and writing, and the checking behind every score never stops.

Outputs are compared against expert human judgment and established CEFR criteria. Trials run continuously. Results are reviewed on an ongoing basis, not just on the day the system goes live. That is the only way to know the technology is still scoring accurately today.

The work is not glamorous. It is the difference between a score you can defend and one that quietly goes wrong for two years.

The Only Way to Scale Language Assessment Without Losing Control

Language assessment has to move forward. The volume, speed, and access that modern hiring and education demand cannot be met by manual marking. Technology is not a threat to quality. It is the only way to deliver quality at scale.

The people do not disappear. Their work changes. They become the validators of what the technology produces, checking it, correcting it, feeding what they learn back so the system performs better over time.

Get that balance right, and no error gets two years to hide.