AI Alignment, Safety and Security

AI alignment concerns the challenge of ensuring that artificial intelligence systems performs in ways consistent with human goals, values, and wellbeing. It spans technical questions of safety, robustness, and reliability, as well as broader ethical and epistemic issues that draw on computer science, psychology, and the humanities.

Mental health is a domain where alignment is especially critical. AI systems are increasingly used in screening, decision support, digital therapeutics, and conversational agents, yet these applications involve sensitive data and vulnerable users. Misalignment, bias, or inappropriate model behaviour can therefore have disproportionately harmful effects.

By focusing on this intersection, the seasonal school provides participants with a timely technical challenge situated in a socially significant context, and offers training in developing AI systems that are robust, reliable, and aligned with human needs.

About A2s2

The A2S2 Seasonal School provides a technically grounded introduction to AI alignment, safety and security, focusing on the whys and hows of making AI systems robust, appropriate, and aligned with human or application goals. While mental health is used as one motivating high-stakes application, the methods, concepts and evaluation frameworks generalize to a wide range of human-centered and safety-critical domains. The seasonal school is aimed to bring together 35-40 PhD, advanced masters students and early-career researchers working in data science, computer science, statistics, cognitive science, psychology, and health informatics with an interest in AI alignment, mental health, and trustworthy AI. The school provides an interdisciplinary environment with participants from both Danish universities and international institutions across Europe. We aim to form a diverse cohort spanning technical, health, and social sciences, and we particularly encourage applications from underrepresented groups.

Dates: 20-23 January 2026 

ECTs: 2.5

Location: La Oficina, Suomisvej 4, 1927 Frederiksberg

Registration details & Fees: Registration 

A2S2 combines lectures, panels, group work, and project-based learning to train early-career researchers to design, critically assess and evaluate aligned and safe AI systems. The school’s broader goals include fostering interdisciplinarity across technical and human-centered fields, strengthening networks between academia, healthcare, policy, and industry, and preparing a new generation of researchers to address urgent challenges in trustworthy AI with direct societal impact.

More details from the official course page: https://kurser.dtu.dk/course/02988

Who should attend

A2S2 is intended for participants working in or entering areas such as:

• AI alignment and safety

• Robust and secure model training

• Large language models and post-training behaviour

• Evaluation, interpretability, and human-centered AI

• Applications in mental health or other sensitive, high-stakes domains

No prior experience with mental health applications is required. 

Program overview

The programme combines technical and conceptual sessions across four themes: 

• Foundations of Alignment and Human Behaviour (Eg: Ultrasocial humans; human judgment in data-driven systems)

• Post-Training and Sociotechnical Alignment (Eg: LLM post-training dynamics; alignment failures; ChatGPT-as-psychotherapist case study)

• Safety, Robustness, and Security (Eg: Robust model training; language model security)

• Evaluation in Sensitive Domains (Eg: Qualitative and quantitative text analysis; mental-health assessment and quality-of-life evaluation)

A detailed schedule can be found here: Program 

Registration Deadline: 05.01.2026

Contact: In case of further queries please reach our to Sneha Das (sned@dtu.dk) & Nicole Lønfeldt (nicole.nadine.loenfeldt@regionh.dk)