The release of ClaudeMythos, where the UK’s AI Security Institute was the only government body outside the US involved, has given this question new urgency.

A detailed concept for a German AI Safety Institute (DAISI) already exists and Tagesspiegel Background has now picked it up. Developed by Kristian Kersting, Kevin Baum and Patrick Schramowski (DFKI), together with Johann Laux (Oxford Internet Institute, University of Oxford), Franziska Boenisch and Adam Dziedzic (CISPA Helmholtz-Zentrum für Informationssicherheit), and Markus Langer (Albert-Ludwigs-Universität Freiburg).

Kevin Baum, Executive Board Member at CERTAIN, explains what it is about: „It is important not to simply build another centre and contribute to fragmentation. A structure must be a roof with interfaces: to standardisation, frontier labs, German industry, science, authorities like Bundesamt für Sicherheit in der Informationstechnik (BSI) and Bundesnetzagentur, but also to politics and administration.”

The guiding principle of the concept: not to slow down, but to enable. Germany is currently one of the few major AI nations without its own safety institute – with geopolitical and technical consequences. Conversations between BSI, the Bundesministerium für Digitales und Staatsmodernisierung and the Bundesministerium des Innern are already underway.

Read the full article in Tagesspiegel Background: https://lnkd.in/dZAs-sYb