Seven percent of teachers in Germany have used artificial intelligence to grade student work without subsequently verifying the results, according to a new national survey. The finding raises fundamental questions about accountability, transparency and the reliability of AI-assisted assessment in education. International data reveal substantial differences in how schools across Europe, the Middle East and Africa are adopting the technology. While AI is increasingly used to prepare lessons and support learning, its involvement in evaluating student performance introduces distinct security, data protection and governance challenges.

Germany: AI-assisted grading raises accountability concerns

Artificial intelligence has become an established part of teaching practice in Germany, but its use is advancing faster than institutional safeguards in some schools. A representative survey published by German digital industry association Bitkom on 27 October 2026 found that 58% of teachers use AI for educational purposes. One in five respondents (20%) reported using the technology to check assignments and examinations. More significantly, 7% acknowledged having used AI to assign grades without checking the results themselves. The findings are based on telephone interviews with 501 teachers working in lower and upper secondary education, conducted between late May and late June 2026.

The survey also identifies considerable uncertainty surrounding the technology. Only 57% of respondents reported having clear institutional rules governing AI use, while 51% said they felt overwhelmed by artificial intelligence. Another 70% believed that students’ use of AI was increasingly difficult to control. Nevertheless, 12% of teachers deliberately permitted AI use in examination situations. Transparency emerged as a central concern: 92% believed students should be required to disclose the use of AI when completing school assignments. The findings highlight a growing imbalance between technological adoption and institutional oversight. While AI is increasingly accepted as a tool for educational preparation, its use in decisions affecting individual academic performance raises questions about professional responsibility and the reliability of automated evaluations.

European schools show sharply different patterns of adoption

International evidence suggests that Germany is not alone in facing these challenges, although comparable figures specifically measuring unchecked AI-generated grades remain limited. The OECD’s Teaching and Learning International Survey (TALIS) 2024, published in October 2025, provides a broader perspective on how teachers incorporate artificial intelligence into professional activities. Across the participating OECD education systems, 36% of lower secondary teachers reported using AI for work-related purposes. However, national differences were substantial. In Romania, Czechia and Malta, approximately 46% of teachers reported using AI. Poland followed closely at 45%, while Austria reached 39%. Adoption was considerably lower in Finland at 27% and France at 14%.

The differences become particularly relevant when examining the use of AI for marking or assessing student work. In Romania, 49% of teachers who already used AI reported applying it to student assessment. The corresponding share was 33% in Poland but only 9% in Czechia and 8% in Austria. These figures demonstrate that extensive AI adoption does not necessarily translate into widespread automated assessment.

AI use in selected EMEA education systems

Source: OECD TALIS 2024, published in October 2025. Figures refer to lower secondary teachers. The first column represents the proportion of all surveyed teachers using AI. The second represents the proportion of AI-using teachers who employ it to assess or mark student work. The figures do not indicate whether teachers reviewed AI-generated assessments.

The distinction between the two indicators is essential. A country may have relatively low overall AI adoption while a comparatively large proportion of its AI users employ the technology for assessment. Likewise, neither indicator establishes how frequently AI-generated grades are accepted without human verification. Germany did not participate in the lower secondary component of TALIS 2024. Consequently, its more recent Bitkom findings cannot be incorporated directly into the OECD comparison.

Middle East and Africa: Adoption levels reveal different priorities

The OECD data also highlight important developments beyond Europe. In the United Arab Emirates, 75% of teachers reported using artificial intelligence for professional purposes, more than twice the OECD average. Among these AI users, 84% employed the technology to research or summarise information, while 82% used it to develop lesson plans or learning activities. Some 38% also used AI to assess or mark student work. Saudi Arabia presented a different pattern. Overall adoption reached 43%, but nearly half of AI-using teachers (49%) reported employing it for assessment. Similar differences appear across African education systems. In South Africa, 48% of teachers reported using AI, and half of those users employed it for marking or assessment. In Morocco, overall adoption was substantially lower at 26%, although 36% of AI users reported employing the technology for student evaluation.

These findings demonstrate that adoption rates alone provide an incomplete picture of AI’s educational significance. A further distinction concerns infrastructure and professional preparedness. Among teachers who had not used AI, 76% in Morocco and 75% in South Africa identified inadequate school infrastructure as a barrier. Such differences are particularly relevant across the EMEA region, where educational institutions operate under widely varying technological, financial and regulatory conditions. For education authorities and technology providers, the implications extend beyond purchasing AI applications. Effective deployment also depends on infrastructure availability, staff training and the ability to maintain appropriate safeguards.

England: Rapid uptake does not necessarily mean automated marking

A separate survey by the UK’s National Education Union provides additional evidence of the distinction between general AI adoption and its use in student assessment. Published in April 2026, the study collected responses from 9,408 union members working as teachers in state schools in England. According to the findings, 76% used AI tools in their daily professional activities, compared with 53% in the previous year’s survey. The most common applications were teaching-resource creation (61%), lesson planning (41%) and administrative tasks (38%). Only 7% reported using AI for marking.

This figure should not be confused with the seven percent identified in Germany. The English survey measured the general use of AI for marking, whereas the German study specifically identified teachers who had accepted AI-generated grades without checking them. The English findings also revealed gaps in institutional governance. According to respondents, 49% of schools had no AI policy covering either staff or students. Although the union survey is not methodologically equivalent to the representative Bitkom study, both investigations point to the need for clearer rules and stronger professional guidance.

Preparing assessments is not the same as assigning grades

Further evidence comes from an April 2026 survey conducted through the European School Education Platform. The investigation collected 1,130 responses from education professionals in more than 50 countries, with particularly strong participation from Italy and Spain. Some 73% reported using AI at least partially to prepare lessons or teaching materials, while 67% used it to develop assessments, including tests, exercises and marking criteria. However, 41% did not use AI for grading or summative assessment at all.

The contrast is significant. Creating an examination question or preparing a marking rubric allows teachers to review and modify the material before it affects students. Assigning grades directly influences individual academic records and potentially future educational opportunities. The European survey was based on voluntary participation and is not representative of all teachers across the continent. Nevertheless, its findings reinforce an important pattern: AI has gained greater acceptance for preparing educational activities than for making final assessment decisions.

The security implications of AI-generated academic decisions

The growing use of artificial intelligence in student assessment raises concerns that go beyond educational methodology. Unlike conventional teaching aids, AI-based assessment systems may process identifiable student information, written assignments, examination responses and historical performance data. Depending on the application and its configuration, this information may be transmitted to external service providers or processed across multiple technical environments. For schools and educational authorities, several risks require particular attention. Data confidentiality and access control are essential where AI applications process personal information relating to children. Institutions need to establish what information is collected, where it is processed, who may access it and whether the processing is legally justified.

Assessment reliability presents another challenge. Generative AI systems may produce plausible but inaccurate evaluations, inconsistently apply marking criteria or misinterpret complex student responses. Bias and fairness also require scrutiny. Differences in language proficiency, writing style, cultural background or the interpretation of a question may influence automated assessments in ways that are difficult to identify. Finally, traceability and accountability become critical when an AI-generated evaluation is challenged. An educational institution should be able to establish which assessment criteria were used, what information informed the evaluation and who approved the final decision. These requirements become particularly important when AI applications are integrated into learning management systems, student information platforms or examination administration environments. In such settings, an inappropriate automated decision may have consequences beyond an individual classroom. The central security requirement is therefore not simply to protect an AI application against unauthorised access. It is also to ensure that its outputs can be verified and that consequential decisions remain subject to effective oversight.

EU AI Act: Educational assessment among the high-risk use cases

The European Union has explicitly recognised the potential risks associated with AI-based educational assessment. Under Annex III of the EU AI Act, systems intended to evaluate learning outcomes in educational and vocational training institutions are among the applications that may be classified as high-risk. The regulation also addresses AI systems used to determine access to educational programmes, assess appropriate educational levels and monitor prohibited behaviour during examinations. High-risk classification depends on the system’s intended purpose and the applicable legal criteria. The use of a general-purpose AI assistant by a teacher does not automatically place every activity within the high-risk category. Where the high-risk requirements apply, the regulatory framework includes obligations concerning risk management, data governance, technical documentation, transparency and human oversight.

Under the amended implementation timetable, the relevant high-risk provisions for standalone systems covered by Annex III are scheduled to apply from 2 December 2027. The AI Act’s human oversight provisions are particularly relevant to automated assessment. They require high-risk systems to be designed so that authorised individuals can understand their limitations, monitor their operation and, where appropriate, disregard, override or reverse their outputs. Existing data protection obligations remain relevant independently of the AI Act implementation timetable. In the European Union, schools processing personal information must comply with applicable GDPR requirements and national education legislation. The regulatory situation differs across non-EU markets in EMEA. Education authorities and system operators must therefore assess the applicable local frameworks rather than assume that a single European compliance approach will cover every jurisdiction.

Teacher competence remains a significant barrier

The international comparison also exposes a skills gap. According to TALIS 2024, approximately 75% of teachers across participating OECD education systems who had not used AI reported lacking the knowledge or skills required to incorporate it into teaching. In France, this figure reached 79%, compared with 69% in Austria and 44% in the United Arab Emirates. These percentages refer specifically to teachers who had not used AI, rather than to the teaching workforce as a whole. The findings demonstrate why access to AI applications cannot be considered sufficient preparation for their responsible use.

Schools also need practical training on evaluating AI-generated content, protecting student information and recognising where automated recommendations may be unreliable. The German Bitkom survey illustrates this tension. While 51% of teachers reported feeling overwhelmed by AI, 50% wanted applications that could help reduce routine administrative workloads. This combination suggests that teachers recognise the potential benefits of automation but require clearer guidance and appropriate skills to use it confidently.

From voluntary adoption to institutional accountability

The available evidence points to an increasingly important distinction in educational AI governance. Using artificial intelligence to summarise teaching materials, generate exercises or prepare administrative documents does not create the same risks as relying on automated systems to determine individual academic outcomes. Nevertheless, many schools are introducing AI into everyday activities before comprehensive institutional policies have been established.

The OECD, Bitkom, National Education Union and European School Education Platform surveys were conducted at different times and used different samples and methodologies. They should therefore not be interpreted as a single statistical ranking of education systems. What they collectively demonstrate is a growing need to distinguish between AI adoption, AI-assisted assessment and the unchecked acceptance of automated decisions. Germany’s latest findings provide an especially important indicator because they address the final category directly. Comparable international data on unverified AI-generated grades remain limited, leaving an evidence gap for policymakers and education authorities. As AI systems become more deeply integrated into educational platforms, future monitoring will need to examine not only how frequently they are used but also how their outputs are validated and how accountability is maintained.

Outlook: The credibility of educational assessment depends on effective oversight

Artificial intelligence offers substantial opportunities to support teachers, personalise learning materials and reduce repetitive administrative work. International data from Europe, the Middle East and Africa nevertheless show that the technology is developing unevenly across education systems, with significant differences in adoption, assessment practices and institutional readiness. The most consequential challenge concerns the boundary between technological assistance and automated decision-making. An AI-generated suggestion can support professional judgement. An AI-generated grade accepted without verification raises a different set of questions about fairness, reliability and responsibility.

For education authorities, cybersecurity specialists and technology providers, this distinction will become increasingly important as AI applications gain access to sensitive student information and established assessment systems. The central question is no longer whether artificial intelligence will influence educational assessment, but whether institutions can ensure that the technology remains transparent, secure and subject to meaningful human oversight.

Sources and methodology

Bitkom (27 October 2026): Seven percent of teachers have awarded grades using AI without verification. Representative survey of 501 lower and upper secondary teachers in Germany, conducted between May and June 2026. OECD (2025):Results from TALIS 2024. International survey of teachers and school leaders, with country-specific data on AI adoption and educational assessment, including Romania, Czechia, Poland, Austria, France, the United Arab Emirates, Saudi Arabia, Morocco and South Africa. European Commission (June 2026): The Teaching Profession in the EU – A Comparative Analysis Building on the TALIS 2024 Results.

National Education Union (2 April 2026): State of Education: AI. Survey of 9,408 teachers working in state schools in England. European School Education Platform (8 April 2026): Survey on Artificial Intelligence for Teaching and Learning – Results. Voluntary survey of 1,130 respondents from more than 50 countries. European Union: Regulation (EU) 2024/1689, Artificial Intelligence Act, as amended in 2026, particularly Annex III and Article 14 concerning high-risk educational applications and human oversight.