Joint enterprises in Britain, the United States and other countries have issued "guidelines for the Development of AI Systems", which is known as "the first relevant safety standard in the world"

CTOnews.com November 28 news, the United States Cyber Security and Infrastructure Security Management Agency (CISA) and the British National Cyber Security Center (NCSC) recently officially released the "secure AI system Development guidelines" (Guidelines for Secure AI System Development) white paper.

The white paper, known as the "first global AI security standard", aims to strengthen the security of AI systems from the design stage to guard against possible security risks.

It is reported that the guidelines are drawn up by 18 countries, including the United States, Britain, Germany, France, Japan, South Korea and Singapore, together with 23 network security organizations, including Microsoft, Google, OpenAI, Anthropic and Scale AI.

▲ source UK relevant departments specifically, the guidelines include four main parts: security design, security development, security deployment and security operations (operation and maintenance). The guidelines propose possible security risks and ways to avoid them for each part, in order to help developers reduce related security risks from the source.

CTOnews.com, the relevant department in the UK, noted that the guidelines mainly provide necessary suggestions for the development of AI systems, and that the development of AI systems should follow the principle of security design (Secure by Design). According to the UK's CISA, the guidelines prioritize customer safety, transparency and accountability, as well as the establishment of an organizational structure for "security design".

This guide applies to all types of AI systems, including not only "cutting-edge models" in the industry, but also other models or API systems that utilize external hosting.