The Biden administration is taking strong measures to ensure the safety and security of artificial intelligence (AI) models by demanding red-teaming safety tests of foundational models. U.S. President Joe Biden will sign an executive order on AI on October 30th, 2023, which will invoke Cold War-era executive powers over private industry. This order will require developers of advanced AI models to notify the government and share the results of safety tests.
The executive order will utilize the Defense Production Act, a statute from 1950, to enforce the notification requirements. Developers of generative AI foundation models that could pose a “serious risk” to national security, national economic security, or national public health will be required to inform the government when training such models. They will also need to share the outcomes of all red-team safety tests.
To ensure the effectiveness of these safety tests, the National Institute of Standards and Technology (NIST) will establish new standards for extensive red-team testing before the public release of a new foundational model. Additionally, the Department of Health and Human Services will create a safety program to address reports of unsafe healthcare practices involving AI.
Bruce Reed, White House deputy chief of staff, described the executive order as the most robust set of actions any government has ever taken on AI safety, security, and trust. Along with the red-teaming safety tests, the order has several other important provisions.
The Department of Homeland Security (DHS) will apply the NIST standards to critical infrastructure sectors and will establish an AI Safety and Security Board. Furthermore, a new government cybersecurity program will be developed to create AI security tools, including the identification and addressing of vulnerabilities in AI technologies.
Federal agencies will be required to adhere to new standards for the use of AI systems and will be encouraged to acquire AI products and services. In collaboration with the Department of Energy, DHS will focus on targeting threats to critical infrastructure that may arise from AI systems, with the aim of developing comprehensive standards to mitigate risks related to chemical, biological, radiological, nuclear, and cybersecurity risks.
In addition to addressing safety and security concerns, the executive order aims to position the United States as a global leader in AI research. To achieve this, grants will be expanded in sectors such as healthcare and climate change, supported by the National AI Research Resource pilot program. This program will provide AI researchers and students across the country access to various resources and data.
Furthermore, the order emphasizes the need for international cooperation in managing the risks associated with AI. The departments of State and Commerce will lead efforts to establish comprehensive international frameworks for AI risk management and safety.
This executive order builds upon previous actions taken by the Biden administration to address AI risks. Last year, the administration published the Blueprint for an AI Bill of Rights, which outlined a framework for the safe and effective deployment of AI and other automated systems. The administration also issued an executive order instructing agencies to combat AI algorithmic discrimination and secured voluntary commitments from 15 AI companies to adhere to standards and requirements for the development of trustworthy AI tools and technologies.
As the Biden administration takes significant steps to ensure the safety, security, and ethical use of AI, the United States aims to lead the way in AI research and development. Through comprehensive standards, red-teaming safety tests, and international cooperation, the government seeks to mitigate risks and build trust in the evolving field of artificial intelligence.