How AI Came to Grade Teachers at Multiverse
Teachers at Multiverse report "horrendous stress" and say they feel constantly watched after the company started using AI to monitor and score their teaching in online classes.
Multiverse, the tech training firm valued at £1.6bn and co-founded by Euan Blair, the eldest son of former UK prime minister Tony Blair, employs more than 800 people and provides apprenticeship training in computing and AI skills to thousands of workers in the UK public and private sectors, according to reporting published 28 September 2026 The Guardian.
The system works from written transcripts. Transcripts of online lessons, meaning word-by-word text records, are analysed and scored by AI that is set to alert human managers when it suspects a teacher has done something wrong. Instructors described the system as "remorseless" and "unnerving."
Each flag is packaged for managers. The system assigns a risk status to each instructor, calculates a percentage confidence score, meaning its own estimate of how sure it is, and writes comments for managers to read. A human review comes after the machine flag. The loop is tight.
That is a change from past practice. Teachers were previously observed once a month by a human manager. They are now checked for several hours per day by AI.
The AI watches both technical handling and speech. It is set to flag teachers who do not fix a lost online connection within about a minute. It is also set to flag filler phrases such as "sort of" and "kind of."
Some staff said they felt "on edge," lost sleep and sought therapy after the monitoring began. Teachers said they felt constantly watched.
Multiverse provides AI and digital skills training to workers in the NHS, local councils, universities and the private sector, mostly paid for through the government's apprenticeship levy, a fund drawn from employers to pay for training. The company uses a mix of AI models, including products from Anthropic and OpenAI, to run its operations and its monitoring system.
The broader context here concerns employment practice and public procurement, and the question is one of proportionality and control. Continuous scoring of transcripts, with risk labels and confidence scores, shifts evaluation from occasional sampling to ongoing audit. For managers, it offers consistency and early warning of delivery problems. For instructors, it turns normal variation in a classroom into machine-readable exception events, with written records for managers created without anyone directly watching.
In my view, the levy-funded part will shape what happens next. When AI and digital training for NHS staff, council officers and university employees is paid through a government training levy, questions about workplace standards, how classroom transcript data is handled, and how much tolerance there is for false alarms carry institutional weight. The choice of model provider matters less than the rules around thresholds, appeal, and how confidence scores feed into performance decisions.
