The ELLIS Institute Tübingen’s Principal Investigators Maksym Andriushchenko and Bernhard Schölkopf contributed to the second annual International AI Safety Report, released in February 2026. Commissioned by the UK Government and spearheaded by Yoshua Bengio, the Report was authored by more than 100 leading experts on AI and provides insights on the latest scientific findings on capabilities and risks of general purpose AI.
Maksym, a leading expert on AI safety and alignment, participated as one of three Chapter Leads, overseeing the sections on current and future capabilities of general-purpose AI. Bernhard was part of the Senior Advisers in 2025, and returned to this position in 2026 as world leading scientist in the field.
The key developments in the field since the release of the annual Report 2025 are, among others, the steadily improving abilities of AI, especially in mathematics, coding, and autonomous operation. This increase in ability comes both from large-scale pre-training and now also from post-training of AI models using techniques such as reinforcement learning with verifiable rewards. The advancing capabilities of AI systems lead to new risks and raise important societal questions that the report discusses in depth.
The Report is led by Yoshua Bengio and supported by a Secretariat within the UK AI Security Institute. He is being advised by an International Expert Advisory Panel, representing 30 countries that participated in the 2023 Bletchley Summit, the European Union, the Organisation for Economic Co-operation and Development as well as the United Nations. The experts work on the AI Safety Report to identify AI risks and technical solutions without suggesting policies, producing a document to help world leaders understand and responsibly treat AI.
Find out more about Maksym’s research.
Find out more about Bernhard’s research.
Learn more about the International AI Safety Report.