The IJCAI-ECAI 2026 conference returns to Germany for the first time in 43 years. On 19 August a representative of the EU AI Office sets out the requirements; an expert panel covers AI agents in public services.
The IJCAI-ECAI 2026 world congress is being held in Bremen – the first time in Germany in 43 years. At its centre sits an expert panel on safe AI agents and the question of how such agents can be operated responsibly in public services.
In parallel the conference addresses implementation of the rules in force since 2 August. On 19 August Jeroen Delfos of the EU AI Office sets out requirements, deadlines and risk classes. That these points are negotiated at a technical congress marks the shift: principles are turning into evidence, documentation chains and approval processes.
The AI festival Bremen Fast Forward follows on 20 and 21 August with around 50 speakers on three stages, with free entry on 21 August for school students, apprentices and university students. For companies the practical part matters more than the programme: with agents, safety emerges from architecture and process decisions, not from a vendor's assurance.
What this means for decision-makers
- For every AI agent in use, record in writing where its scope of action begins and ends and who signs off results.
- Build an evidence chain: which inputs were checked, which outputs were used, who decided.
- Use the EU AI Office's guidance as the basis for your internal approval rules rather than waiting for national clarifications.
This story was produced automatically from the source named above and checked by software before publication. The image is symbolic and shows neither the event nor a real person. How this paper is made