EPSO AD8 in artificial intelligence: EPSO/AD/430/26

EPSO/AD/430/26, published on 8 September 2026 (Official Journal C/2026/4668), builds reserve lists of AD8 administrators in two fields. Field 1 is artificial intelligence, with 240 places; field 2 is cybersecurity, with 254 and its own guide. You apply for one field only.

It is aimed at experienced professionals: AD8 is above the usual entry grades, and every route to eligibility asks for at least five years of relevant experience. The notice also makes clear that the profile is technical: every category of duties "requires technical expertise", whether the post is about building AI systems, governing them or advising on AI policy.

Besides EU citizenship and two EU languages (one at C1, another at B2), you need one of four combinations of studies and experience. The experience counts from the date of your diploma and has to match the AI duties in Annex II of the notice.

A master's or a PhD in an accepted area satisfies the four- and five-year options whatever your first degree was.

The accepted areas are AI, computational linguistics, data science, computer science or IT, electrical or electronics engineering, telecommunications, software engineering, cybersecurity, cloud computing, high-performance computing, quantum technologies, information systems, statistics and mathematics. Any other discipline also counts if your studies had a focus on AI, proven by a recognised AI specialisation on the diploma or transcript, at least 30 ECTS of AI (one semester full time), or a thesis on AI.

All tests are remote and supervised, on the dates in your invitation, and not necessarily in this order.

The reasoning tests have the same format and pass marks as in EPSO's other 2026 competitions. For the platform and the rules of remote testing, see the guide to TAO.

In a strict order, set out in the notice:

1. The reasoning tests are marked first. Fail them and nothing else is marked.
2. The AI field test is marked for those who passed reasoning. Everyone with at least 15/30 is ranked by that score.
3. From the top of that ranking, a limited number of candidates, in principle no more than 1.5 times the places, have their essay marked and their eligibility checked. In AI that is about 360 candidates, plus any ties at the cut-off.
4. Those who pass the essay and are eligible go onto the reserve list, in ranking order, until the 240 places are filled.

So the AI test is the only score that counts. There is no interview and no assessment centre before the list, and results are only sent at the end of the competition, whatever stage you reached.

The notice publishes no syllabus: the test is "specific to the field". The closest thing is Annex II, which sets out three categories of duties. It is also what your experience is measured against.

1. Building and running AI solutions end to end. Advising on architectures, algorithms and design patterns at enterprise scale, build or buy decisions, and model selection and optimisation; leading AI projects into production; the foundations of AI at scale: system security, regulatory compliance, full MLOps, the product lifecycle, testing and quality assurance; data readiness: preparation, pipelines, quality, robustness and governance; and watching generative AI models, APIs, cloud, integration technologies, RAG, protocols such as MCP, and agentic AI.
2. Compliant, defensible and scalable use of AI. AI governance frameworks and risk management; compliance with the AI Act, internal AI policies, data protection and security, including tools and benchmarks to evaluate model capabilities and risks; trustworthiness, reliability, auditability, fairness and safety before deployment and in use; explainability, traceability and reproducibility; building compliance into projects and contracts; and AI literacy.
3. AI policy and its enforcement. How AI technology meets policy domains; trends and societal impact; watching the general-purpose AI ecosystem and new risks; assessing the AI strategies of public and private organisations; and making policy decisions technically sound.

Knowing the AI Act is not enough, and neither is knowing how to train a model: the annex puts the three categories side by side. The rules worth having at hand are the AI Act (Regulation (EU) 2024/1689), with its risk-based obligations and its rules for general-purpose AI models, and data protection: the GDPR and, inside the institutions, Regulation (EU) 2018/1725.

1. Check your eligibility now. Count your years of experience from the date of your diploma. If your degree is not in the list, gather the proof of your AI focus: transcript, ECTS or thesis. Documents are due by 14 January 2027, but a gap found then cannot be fixed.
2. Do not underestimate reasoning. It only has to be passed, but it comes first, and an experienced professional who has not sat a psychometric test in years can fail it with a CV that would top the ranking. Practise the real format: 40 questions in 65 minutes across three timed tests.
3. Put most of your time into the AI test. It is the only score that ranks you: 30 questions in 40 minutes, about 80 seconds each, in your second language. Cover the three categories of Annex II, especially the one furthest from your own job.
4. Learn the vocabulary in your second language. MLOps, data governance, explainability or risk management will come to you in that language, not in the one you work in.
5. Read the essay documentation as soon as EPSO publishes it. The EUFTE is based on EU documentation that EPSO puts on its website before the test, and it is marked on written communication, not on facts or language.
6. Apply before the last day. The deadline is 13 October 2026 at 12:00 Brussels time, and the ID copy must be uploaded by then too.