AI in Marking
As 1.6 million students across India (and abroad) grapple with their CBSE class 12 results and the use of On Screen Marking (OSM), the impact of technological change in high-stakes assessment – and the regulator’s role in approving it – is in the spotlight once again. Students and parents have been raising concerns that the grades are lower than expected and many are attributing them to the newly introduced OSM. It invites an interesting comparison with what is happening elsewhere in the world, particularly in the UK, where OSM for high-stakes examinations is the norm, but the debate is more about use of AI in those evaluations than about OSM itself.
On Screen Marking (OSM) has been used by all four English awarding organisations in the UK for assessing GCSEs over many years. Each awarding organisation has implemented OSM through a rigorous change programme for the rollout, including multiple pilots before moving to full-scale operation. The change programme involved optimisation of the processes, development of the platform in line with the revised process and multiple runs – including non-live and live pilots. Considerable effort has been spent on testing the performance of the system – starting from the scanning of the papers all the way up to result publishing – before using it to decide the fate of students. Each awarding organisation has given ultimate priority to the quality control system which ensures integrity, accuracy and fairness of the results for each candidate.
Despite all these efforts, most years see the awarding organisations go through various challenges – such as system outages, performance issues and candidate appeals – which are handled through standard operational processes (IT Service Management, for example). We hope that, when the dust settles, CBSE will look at the lessons and fix the problems, rather than reverting to manual marking. Any change programme that touches the lives or livelihood of people will trigger public anxiety. If the objective is fair and clear, it is up to the organisation to ‘fix forward’ rather than ‘regress’. What matters is clarity of direction.
And on that, Ofqual’s chief regulator Sir Ian Bauckham has been direct in his report in Tes on 14 January 2026, alongside Ofqual’s working paper ‘Principles of AI use in marking’. Bauckham clearly set out where AI currently sits in the marking process. “AI is promising for Quality Assurance and Marker Training,” he wrote, “but for the moment it’s nowhere near ready to take over high-stakes marking.” Ofqual considers sole AI marking non-compliant with current regulations, not merely premature. Notably, the objection is to the role of AI in marking – with or without humans in the loop – and not in quality assurance, which is essentially the same activity but with a different sequence of events. The awarding organisations have openly piloted the quality assurance model in previous years with varying degrees of success.
There are two elements to the problem we are trying to solve. As most of the high-stakes examinations are paper-based today, the first element is converting handwritten pages to machine-readable format. And the second element is assessing these against an agreed mark scheme. Both have challenges of their own, and both are prone to bias. Despite the significant improvements over the years, the conversion of handwritten papers arguably remains a substantial hurdle. Ofqual’s on-screen consultation, which ran from 11 December 2025 to 5 March 2026, proposes that each awarding body could introduce up to two new on-screen specifications, with subjects above 100,000 entries excluded for now. If successful, this could eventually transform GCSE delivery into a fully digital model — at which point the handwriting recognition problem largely disappears, and AI becomes a more viable option for the marking itself.
For now, the interesting question is what awarding bodies are doing in the gap between what the regulator forbids and what it permits. Bauckham’s line is that AI cannot be “the” marker. It says nothing about AI being “a” marker – for example, a flagging mechanism. That role is not just permitted; it is being actively encouraged.
The challenge is now for the technology partners, who carry significant responsibility in implementing the awarding organisations’ vision and objectives, to work closely with them to come up with a model that works best. In doing so, they will have to deliver a careful rollout, without pupils and parents having to wonder whether they got the raw end of the deal.