Ceibal & ANEP's AI-Assisted Basic-Cycle Accreditation Exam

Uruguay's Ceibal agency, working with the national education authority ANEP, built an AI grader for the country's basic-cycle accreditation exam — a route back into the credentialed system for young people and adults who left school early. Around 8,000 candidates were enrolled to sit the exam on 13 June 2026.

The exam has three parts: multiple-choice reading comprehension, multiple-choice mathematics, and a written essay scored against 15 specific criteria such as organisation, spelling and argumentation. Ceibal engineers spent more than a year building, testing and calibrating the system against prior years' human-scored results, using a contracted OpenAI model wrapped in Uruguayan-built software. On the writing section, the AI auto-scores where confident, flags a human corrector when unsure, and steps in as a neutral 'third reader' when two human correctors disagree; it can also be used by candidates as a self-study tutor before the exam.

Ceibal reports that in testing, the AI's essay scores matched human correctors' assessments, and the system does not store student data or access ANEP's own student records, linking only through anonymised ID numbers.

This validation is described in press coverage as an internal testing-phase comparison rather than an independently published psychometric study, and the tool is so far deployed for a single exam track within a country whose Ceibal programme already gives 100% of public grade 1-9 students a personal device.