AI Evaluations · Model Monitoring · Applied Machine Learning

At the European Commission AI Office, I work on emerging risks from general-purpose AI, including harmful manipulation, child safety and post market monitoring. I've been working on AI for +8 years, and my broader research and work experience spans the whole AI life cycle, from data quality to deployment and monitoring and policy. I also lead multidisciplinary technical teams and external research collaborations, and have published in Science, Nature Machine Intelligence, NeurIPS, AAAI, AIES and TMLR.

Before joining the European Commission, I was a Principal Investigator and Applied Skills Advisor at The Alan Turing Institute a Marie Skłodowska-Curie Research Fellow Marie Curie Action , a statistician at the European Central Bank European Central Bank , and a consultant at Deloitte Robotics Deloitte. I also held research positions and visits at BBC, Barcelona Supercomputing Center Barcelona Supercomputing Center, IIIA-CSIC IIIA-CSIC, and SCHUFA. Before my research career, I competed as an elite swimmer Swimming.

LinkedIn ↗ Google Scholar ↗
Carlos Mougan
Currently
Technical

AI evaluations & model behaviour

Design and lead evaluations of frontier AI models, translating ambiguous questions about model behaviour into measurable experiments. See our Science paper on AI evaluations for an example of this work.

Research

Emerging AI risks

Research emerging behaviours and risks from frontier AI, including post-market monitoring, manipulation, persuasion, over-reliance, child safety and mental-health interactions.

Applied

Technical projects & mentoring

Lead multidisciplinary technical teams and external research collaborations on post-deployment evaluation and monitoring, building on my PhD research on detecting changes in model behaviour after deployment.

Experience

Applied research, machine learning and policy

I've worked across research, technical projects and public institutions, combining applied research with product.

2024 — Present

AI Office — AI Safety

European Commission · Brussels

Design and lead technical evaluations of general-purpose AI models, including experimental work on manipulation, persuasion, child safety and post-deployment monitoring. Develop evaluation methodologies, coordinate external technical research teams, and monitoring emerging behaviours after deployment.

2022 — 2025

Principal Investigator, Applied Skills Team

The Alan Turing Institute · London

Technical and research advisory work for public, nonprofit and industry partners such as: British Antarctic Survey, WWF, Amnesty International, Roche, Astra Zeneca, Defence and Security Gov, Environmental Investigation Agency, John Muir Trust, CEFAS, Telenor, Turkcell, ...

2021 — 2024

Marie Skłodowska-Curie Research Fellow & PhD Researcher

University of Southampton

Research on model monitoring, explainability and fairness within the EU-funded NoBias Innovative Training Network, including research visits at SCHUFA, the University of Pisa and BBC DataLab.

2020 — 2021

Statistician, Directorate General Statistics

European Central Bank · Frankfurt

Statistical quality monitoring across EU Member States and applied machine-learning work for the European System of Central Banks.

2017 — 2018

Technology Consultant

Deloitte Robotics · Madrid

Applied OCR and robotic process automation to audit and banking workflows, and trained Deloitte teams on UiPath.

Research

Selected publications.

My research spans the full AI lifecycle: data collection, data quality, preprocessing, modeling, and monitoring. I have led publications to NeurIPS, AAAI, AIES, TMLR and Science.

2026 · Science
The science and practice of proportionality in AI evaluations
Carlos Mougan, Lauritz Morlock, et al.
Science
2023
How to Data in Datathons
Carlos Mougan, Richard Plant, Clare Teng, Marya Bazzi, et al.
Advances in Neural Information Processing Systems (NeurIPS)
2023
Monitoring Model Deterioration with Explainable Uncertainty Estimation via Non-parametric Bootstrap
Carlos Mougan, Dan Saattrup Nielsen
AAAI Conference on Artificial Intelligence
2023
Fairness Implications of Encoding Protected Categorical Attributes
Carlos Mougan, Jose Alvarez, Salvatore Ruggieri, Steffen Staab
AAAI/ACM Conference on AI, Ethics, and Society (AIES)
2021
Quantile Encoder: Tackling High Cardinality Categorical Features in Regression Problems
Carlos Mougan, David Masip, Jordi Nin, Oriol Pujol
Modeling Decisions for Artificial Intelligence
2021
Desiderata for Explainable AI in Statistical Production Systems of the European Central Bank
Carlos Mougan, Georgios Kanellos, Thomas Gottron
ECML PKDD Workshops

View all publications on Google Scholar ↗

Background

Education and selected highlights.

Education

PhD — Model Monitoring & Machine Learning University of Southampton · 2021–2024 Thesis: Model Monitoring in the Absence of Ground Truth Data via Feature Attribution Explanations.
MSc — Mathematical Modelling Universitat Autònoma de Barcelona · 2018–2019
BSc — Physics Universidad Complutense de Madrid · 2013–2018

Other highlights

  • Elite swimming athlete, 2013–2017
  • Main contributor to Category Encoders; main developer of skshift and explanationspace.
  • Ranked among the top two Data Science Stack Exchange users in 2020.
  • 1st @BCG Gamma Datathon; 1st @Novartis 2021 Datathon; Bronze Medal @ Kaggle IEEE-CIS Fraud Detection.
Selected talks

Presentations & invited talks.

Royal Society for Spanish Mathematics · 2024 Beyond Demographic Parity: Redefining Equal Treatment
NeurIPS · 2023 How To Data in Datathons
Regulatable ML Workshop, NeurIPS · 2023 Necessity of Processing Sensitive Data
AAAI · 2023 Monitoring Model Deterioration with Explainable Uncertainty Estimation via Non-Parametric Bootstrap
Outside work

Fun pictures

Board Kitesurfing At The Alan Turing Institute Chile