The United States, Britain, Australia and 15 other countries have issued global guidelines to help protect artificial intelligence models from tampering, urging companies to make their models "safe by design." On November 26, the 18 countries released a 20-page document outlining how artificial intelligence companies should handle their cybersecurity issues when developing or using artificial intelligence models, as they claimed that "security is often a secondary consideration" in this fast-paced industry.
The guidelines mainly include general recommendations, such as tightly controlling the infrastructure of AI models, monitoring models before and after they are released, and training employees on cybersecurity risks.
But certain controversial issues in the field of artificial intelligence are not mentioned, including possible controls on the use of image-generating models, deepfakes or data collection methods and their use in training models - an issue where several artificial intelligence companies have been sued for copyright infringement.
U.S. Department of Homeland Security Secretary Alejandro Mayorkas said in a statement: "We are at an inflection point in the development of artificial intelligence, which may very well be the most influential technology of our time. "Cybersecurity is the key to building safe, reliable and trustworthy artificial intelligence systems."
The guidelines follow other government-to-government initiatives to weigh in on AI, including an AI safety summit held by governments and AI companies in London earlier this month to coordinate agreements on AI development.
Meanwhile, the European Union is working out the details of its artificial intelligence bill that will provide oversight of the field, and U.S. President Joe Biden issued an executive order in October setting out standards for AI safety and security - although both orders were opposed by the AI industry, which claimed they could stifle innovation.
Other co-signatories of the new "Safe by Design" guidelines include Canada, France, Germany, Israel, Italy, Japan, New Zealand, Nigeria, Norway, South Korea and Singapore. Artificial intelligence companies including OpenAI, Microsoft, Google, Anthropic and ScaleAI also contributed to the development of the guidelines.