The London School of Innovation became the first UK institution to receive regulatory approval for AI-taught master’s degrees in March. Tuition costs between £9 000 and £11 000 for a one-year online programme delivered primarily through AI avatars running Socratic dialogues. Enrolments start in June.

The relevant question for a tech-industry reader is which version of AI-in-education survives this cycle. CambriLearn has spent 20 years building the answer.

CambriLearn is an accredited international online private school. With close to two decades of operation, it has taught more than 80 000 students across more than 100 countries, in five curricula, namely British, Pearson Edexcel, Caps, IEB and US K-12.

It runs live, time-tabled lessons with qualified specialist teachers, and its accreditation and registration coverage spans Cognia, Pearson Edexcel, SACAI, IEB and NCAA. The school is not an ed-tech vendor, and that distinction governs how it deploys technology, including AI.

AI in K-12 is the most consequential technology decision the schooling sector will make this decade, and most operators are getting it wrong. Pew Research data from late 2025 shows 54% of US teenagers already use AI chatbots for schoolwork. Securly’s analysis of 1.2 million student-AI interactions across more than 1 300 US school districts between December 2025 and February 2026 found that roughly one in five involved problematic behaviour, including cheating, bullying and self-harm content.

AI tutors

The Brookings Institution’s January 2026 report on AI in education concluded that under current deployment patterns, the risks to K-12 students outweigh the benefits. The commercial corollary is India’s ed-tech correction. For example, Byju’s AI-first product positioning, layered onto thin pedagogical foundations, did not survive contact with actual children.

The harder problem CambriLearn is building for is AI deployed as a tool inside a teaching operation. That is a more demanding engineering brief than building an AI tutor. It requires the AI to operate inside a structure where a qualified teacher is responsible for the child, where the curriculum is governed by external examination standards and where the technology has to integrate with 20 years of teaching practice rather than displace it. AI tutors built outside that constraint can ship faster, and they produce the failures the K-12 data is now documenting.

CambriLearn is using AI inside the school today, and the application is widening. While most schools are still drafting their first AI policies, CambriLearn is already deploying AI tools inside its operations, with more rolling out over time.

The deployment is governed by the same two questions every tool has had to answer for 20 years: does it help the teacher teach better, and does it help the child learn better? Tools that pass both are kept and extended. Tools that fail either are discarded. That discipline is what distinguishes a school running AI from a technology product running children.

Read: AI-ready schools already exist – just not in physical classrooms

The structural point matters for anyone reading this from the operator side. Schools that survive this cycle will be the ones with deep teaching IP sitting inside the same operation as the engineering function. Operators that licensed a model and built a UI on top will not. CambriLearn’s 20-year operating history, across five curricula and 100-plus countries, is the kind of institutional foundation an AI deployment in K-12 actually requires, and is not something a competitor can replicate by raising a series-A round.

Read: What South African parents look for in an online school

The London approval of AI-taught master’s degrees will generate plenty of headlines. The more useful test is which schools are still here in five years, still producing measurable outcomes for children. CambriLearn has been answering that test in production, at global scale, for two decades, and AI is the next chapter in a discipline the school has been practising the whole time.