A company that trains workers in artificial intelligence skills is now using artificial intelligence to monitor the people doing the training. That tension sits at the heart of a recent whistleblower account from Multiverse, a UK tech training firm valued at £1.6 billion, where instructors say a newly introduced AI monitoring system has left them losing sleep, seeking therapy, and questioning every word they say in the classroom.
A System That Never Looks Away
Before the new system arrived, Multiverse instructors were observed by a human manager roughly once a month. Now, AI models analyse transcripts of their online sessions for several hours each day. The shift is not just quantitative. It is qualitative.
The AI assigns each instructor a “risk status” and a “percentage confidence score,” then generates written comments for managers to review. It flags specific behaviours: using filler phrases like “sort of” or “kind of,” failing to resolve a connection disruption within approximately one minute, giving vague answers to student questions, or getting drawn into a prolonged one-to-one exchange that sidelines the rest of the group. If an instructor self-corrects while giving instructions, that self-correction is itself counted as evidence of poor preparation, even if the explanation was eventually clarified.
The system is instructed to look for patterns across entire transcripts, not isolated moments. Its own documentation distinguishes between occasional hedging, which it tolerates, and what it calls a “pervasive repeated pattern,” which it penalises. The logic sounds measured on paper. In practice, instructors say it has produced something closer to paralysis. One described no longer being able to focus on the learner in front of them, but instead monitoring their own speech in real time, asking themselves whether the AI will flag what they just said.
The Gap Between the Company’s Framing and the Workers’ Experience
Multiverse’s response to the controversy is worth examining carefully, because it illustrates a pattern common to AI deployment in workplaces. The company states that the AI does not replace human judgment but “directs human time to where it’s needed most.” Only human managers write formal performance reviews. The most serious action the system can take, according to a spokesperson, is to recommend that a human personally reviews a session.
This framing positions the AI as a triage tool, a filter that surfaces problems so that human attention can be applied efficiently. It is a reasonable description of what the technology does mechanically. What it does not address is what continuous automated scrutiny feels like from the inside.
Several instructors, speaking anonymously, described the experience as “remorseless” and “unnerving.” One said the stress had disrupted sleep. Another said the system focuses attention entirely on what was done wrong, creating a mental environment of perpetual anticipation of failure. A third described being flagged by managers for apparent “digressions” that were, in fact, responses to student questions: the AI had misread adaptive teaching as deviation from the script.
There is also a transparency problem. During the rollout, instructors were not given access to the AI’s assessments of their own sessions. Managers could see the dashboard; teachers could not. The company says access will be provided in time. But the asymmetry during the introduction period meant workers were being evaluated by a system whose outputs they could not see or contest.
Multiverse uses a mix of AI models including products from Anthropic and OpenAI. The company was co-founded by Euan Blair, who holds a large minority stake and whose estimated personal net worth stands at £350 million. The firm reported a turnover of £118 million and losses of £28 million in its most recent financial year. In February, Ofsted, the UK education inspectorate, found Multiverse failed to meet expected standards in several areas, while specifically praising its instructors as “skilled, effective teachers.”
What This Case Reveals About AI at Work
Multiverse is not alone. Burger King has used AI to track customer interactions in restaurants. Meta paused a programme monitoring employee keystrokes following data privacy concerns and internal pushback. The pattern is consistent: AI monitoring tools are being introduced across sectors, often faster than the frameworks for managing their human consequences.
The Multiverse case is particularly pointed because of what the company does. It trains workers in AI and digital skills, largely funded through the UK government’s apprenticeship levy, serving the NHS, local councils, universities, and private sector employers. The instructors raising concerns are themselves teaching the workforce of the future how to work alongside AI. The irony is not subtle.
What the case surfaces is a distinction that often gets lost in discussions about AI in the workplace. There is a difference between AI that augments human judgment and AI that substitutes for the texture of human observation. A monthly review by a manager who knows the classroom context, the particular learner group, the unexpected question that derailed a plan, carries information that a transcript score cannot capture. When the score becomes the primary signal, something is lost, not just in worker wellbeing, but in the quality of the assessment itself.
The Communication Workers Union’s national tech officer called for new regulation to ensure AI is deployed in socially responsible ways. That call reflects a broader gap: the technology is moving faster than the governance structures designed to manage it.
In Short
AI monitoring of workers is not new, but the Multiverse case makes the stakes unusually visible. A system designed to improve teaching quality is, by the accounts of the teachers themselves, degrading the conditions in which good teaching can happen. The tool flags filler words and self-corrections. It cannot flag the cost of making every instructor afraid to think out loud. That gap between what an AI can measure and what actually matters is the real lesson here.
Based on reporting from The Guardian - Technology.