US Finalises Voluntary AI Cybersecurity Testing for Advanced Models

The US government has finalised its voluntary testing regime for advanced artificial intelligence models, focusing on critical cybersecurity risks. This program emerged from an executive order issued by President Biden in October 2023. The initiative aims to develop robust safety practices for AI systems.

Specific attention is directed towards dual-use foundation models. These models possess capabilities that could pose a serious risk to national security, national economic security, or national public health and safety. The testing framework addresses potential vulnerabilities within these powerful AI systems.

The Department of Homeland Security (DHS) leads the development of an AI Safety and Security Centre. This centre will coordinate efforts across government to assess and mitigate risks posed by AI. It will also foster collaboration with the private sector to share best practices.

The National Institute of Standards and Technology (NIST) is tasked with creating standards for red-teaming AI models. These standards will guide evaluations of AI systems for harmful capabilities or emergent properties. Such assessments are crucial for identifying unforeseen risks before deployment.

These voluntary tests represent a foundational step. They seek to establish a baseline for AI safety and security practices. However, the voluntary nature raises questions about universal adoption across the industry.