Germany is setting up a national artificial intelligence (AI) safety institute to better assess the opportunities and risks linked to advanced AI models.
Digital Minister Karsten Wildberger said on Tuesday that the new body – modelled on the British institute – should bring together “top expertise from world-leading experts” to quickly test new models such as Anthropic’s Claude Mythos and offer advice to the German government.
The new German institute was approved by the National Security Council on Monday, aiming to analyse the capabilities and potential dangers of AI systems.
The institute is also expected to work more closely with similar bodies abroad and push for common international standards on the use of AI.
Key aims of the new body include a focus on AI security – meaning protecting against cyberattacks – but also on AI safety, ensuring models do not become dangerous for humanity.
The development of Mythos made waves earlier this year after Anthropic warned the new model could uncover software vulnerabilities in operating systems, leading the company to delay its release.
While the German government initially did not receive access to the new tool, the Federal Office for Information Security (BSI) – which has led the government’s efforts on AI security so far – is in contact with Anthropic.
The new German AI agency is to begin work as a virtual institution, before a site is found as a suitable headquarters.
Many countries already have such institutes. An international meeting on standardizing AI safety standards was held in Seoul in December, with the BSI also taking part.
Germany’s coalition government set up the National Security Council last year to respond quickly and effectively to crises and threats of all kinds.
The council, chaired by German Chancellor Friedrich Merz, also discussed the security situation in the Middle East on Monday.
It decided to review laws governing the supply of essential goods and critical infrastructure to the population and the armed forces in times of crisis, and adapt them to current challenges.