Whitepaper Frontier AI Safety: Establishing Effective Governance Structures and an International Testing Framework

In a joint white paper, TÜV AI.Lab and the TÜV Association present a proposal for establishing an independent safety infrastructure for advanced AI models. Germany and Europe must be able to independently assess the safety of frontier AI.

© Curated Lifestyle via Unsplash

Highly advanced AI models (frontier AI) and autonomous AI agents offer significant opportunities but also pose novel security risks. Existing regulatory approaches and voluntary commitments by manufacturers are not sufficient. In a joint white paper, TÜV AI.Lab and the TÜV Association call for independent testing and the establishment of an internationally binding security framework.

The White Paper at a Glance

  • Independently test frontier AI: Advanced AI models must undergo independent testing throughout their entire lifecycle. Germany and Europe must be able to assess the risks posed by these systems, even if they were developed outside Europe.
  • Expand the German AI Safety and Security Institute (AISI): AISI should become a centre of competence for frontier AI and national security. This requires two pillars: First, the capability to independently evaluate and test frontier models throughout their entire lifecycle. Second, expertise in safety, specific domains and testing is needed when frontier AI interacts with critical infrastructure, industrial facilities, vehicles or other systems in the physical world. AISI should therefore be established in cooperation with independent testing organisations from the outset, systematically integrating their safety and domain expertise built up over decades.
  • Safeguard the European AI regulatory framework: Consistently implement a coherent EU regulatory framework and align national testing and supervisory structures with European requirements.
  • Expand international cooperation: The EU should work with like-minded countries to develop internationally harmonised safety criteria and testing mechanisms.
  • Establish a binding global security framework: Introduce internationally binding safety requirements as well as coordinated testing mechanisms, reporting procedures and supervisory structures.

Download

Whitepaper Frontier AI Safety: Establishing Effective Governance Structures and an International Testing Framework