Companies that design powerful artificial intelligence systems must perform safety tests on the programs and share results with the U.S. government under a sweeping executive order intended to make AI safe while furthering robust innovation.
The order responds to calls from Big Tech, consumer advocates, and others for guardrails about how AI can and should be used, especially generative AI. Congress has held numerous hearings and a well-publicized forum about AI but has yet to move ahead with comprehensive legislation. Meanwhile, the European Union is expected to finalize its AI Act by the end of the year.
Under President Joe Biden’s order, the National Institute of Standards and Technology (NIST) will create the tests companies can use to assess AI’s risks, as well as standards and other tools. NIST has been focused on AI for years; in January, it released an AI risk management framework believed by some to be a forerunner to federal regulations to curb AI risks among developers and users.
Companies creating AI programs that could pose a serious national security risk or endanger public health must alert the Department of Homeland Security (DHS) before they begin testing their systems, the Biden administration said. The DHS will create an AI safety and security board focused on the risks of AI to critical infrastructure, cyber, chemical, nuclear, and other areas.
“Together, these are the most significant actions ever taken by any government to advance the field of AI safety,” the administration said.
The Department of Commerce will draft guidance for “watermarking” any content from the government that was created with AI to verify authenticity. The government wants the guidance to also serve as a nudge to the private sector to label AI-generated content.
The government will further build new AI tools to find and fix vulnerabilities in critical software to make them more secure.
Biden said his administration will provide support to create better privacy protections around AI, including by funding a research coordination network to advance new privacy tools.
The administration will issue guidance to federal contractors and landlords to ensure AI algorithms in use do not exacerbate discrimination. The Department of Justice will prosecute companies that violate civil rights through AI.
The Department of Labor will address job losses because of AI.