Germany plans an AI Safety Institute modeled on the UK approach

Germany’s National Security Council has approved plans for an AI Safety Institute modeled after the UK’s AISI, according to The Decoder. The institute is expected to test frontier models from companies such as Anthropic and OpenAI for security risks.

The move matters because AI safety evaluation is becoming a state-level capability, not just an internal lab function. Governments want independent ways to assess risks from the most capable systems.

The harder strategic issue is dependency: Europe still relies heavily on US and Chinese frontier model providers, which limits how much control a national institute can exercise over the technology it evaluates.