Skip to main content

President Biden Signs United States’ First Executive Order on Artificial Intelligence, Addressing Safety and Security, Competition, and Civil Rights

Client memorandum | November 3, 2023

Authors: Amir Ghavi, Richard Powers, Katelyn E. Katsuki

On October 30, 2023, President Biden signed the first executive order (“the Order”) in the United States regulating the use of artificial intelligence (“AI”).[1] The Order follows previous steps taken by the Biden administration to address AI development and guidelines, including the voluntary commitments from leading AI companies secured over the summer.[2] Characterized by the White House as the “most sweeping actions ever taken” by a government to promote the advancement of AI safety, the Order represents the first binding AI-specific federal government action in the United States and amounts to a series of agency-specific studies on AI implementation. In other words, the Order does not commission a study on whether the U.S. government should implement AI — it assumes that AI is a crucial technology and creates a framework for how to best integrate its use in our society. The Order tasks various federal agencies to implement a number of policies and procedures and establishes the White House Artificial Intelligence Council (“White House AI Council”) to coordinate those agencies’ activities. Through its guiding principles and priorities, the Order: (i) requires builders of sophisticated AI models to share safety test results with the U.S. government before releasing their systems,[3] prioritizing the development of standards for stress-testing AI systems and watermarking AI-generated content; (ii) directs federal offices to develop standards for investigating and prosecuting civil rights violations related to AI, aiming to prevent hiring, housing, credit and other similar discrimination; and (iii) encourages federal agencies to promote competition in AI and related technology markets, among others. 

In a coordinated action, the Order was delivered on the same day that Japan announced details of the Hiroshima AI Process, an international set of guiding principles and voluntary code of conduct endorsed by the G7 targeting companies involved in the development of AI technology.[4] The Biden administration announced that it has consulted widely on AI governance over the past several months with allies and global partners, and that the Order aligns with ongoing discussions held at the United Nations. Additionally, the Order was released days ahead of the UK AI Summit, where Vice President Harris announced a series of new initiatives focused on the safe and responsible use of AI, including the creation of the AI Safety Institute, the issuance of draft policy guidance on the U.S. government’s use of AI, and the launch of the New Funders Initiative to advance AI in the public interest.

I.  New Safety and Security Standards for Artificial Intelligence[5]

The Order emphasizes that AI must be safe and secure prior to its use.  That requires robust, reliable, and standardized evaluations of AI systems to test and mitigate risks associated with the use of AI systems. The Order mandates the following:[6]

  • Disclosure of Safety Test Results. Starting in January 2024, companies creating powerful foundation models (i.e., those with computational power well beyond that of current commercially available models)[7] must report to the federal government on an ongoing basis any continuous or planned activities related to the training, development, and production of dual-use foundation models,[8] ownership and possession of model weights, and the physical and cybersecurity measures taken to protect the training processes and the model weights from sophisticated threats. Companies must also provide the outcomes of all red-team safety assessments.[9] Additionally, companies or individuals that possess, acquire, or develop a potential large-scale computing cluster must report such facts including the existence and location of the clusters and the amount of total computing power available in each cluster.[10]

    • By January 2024, the Secretary of Commerce must also propose regulations that mandate cloud computing providers(known as IaaS) [11]  to report when a foreign individual or entity attempts to train a powerful AI model.[12] By April 2024, the Secretary of Commerce is required to establish standards for IaaS providers with respect to foreign resellers of their IaaS services, including documentation and identity verification procedures, recordkeeping requirements, and methods to limit third-party access to information.

  • Development of Standards and Practices to Detect AI-generated Content. By June 2024, the Department of Commerce is required to submit a report to the Director of the Office of Management and Budget (“OMB”) identifying standards and methods for content authentication and watermarking to effectively label AI-generated content, detect synthetic content, and prevent generative AI[13] from producing child sexual abuse material or non-consensual intimate images of real individuals. Within 180 days of submitting the report, the Secretary of Commerce is required to develop guidance regarding the tools and practices for digital content authentication and synthetic content detection measures.

  • Soliciting Input on Dual-Use Foundation Models with Widely Available Model Weights (i.e., Open Source Models). By July 2024, the Secretary of Commerce is required to seek input from the private sector, academia, civil society and other stakeholders on and assess the advantages and risks associated with widely available weights for dual-use foundation models,[14] submit a report to the President outlining the potential benefits, risks, and implications of dual-use foundation models with accessible weights, and provide policy and regulatory recommendations concerning these models. This means that, unlike the EU AI Act, which has sought to regulate open source software through a series of revisions and a general lack of public debate — the U.S. government is encouraging a debate involving myriad and impassioned voices surrounding open source software.

  • Development of Standards, Tools, and Tests to Ensure Safe, Secure, and Trustworthy AI Systems. By July 2024, the National Institute of Standards and Technology (“NIST”) must establish guidelines and best practices for the development and deployment of safe, secure, and trustworthy AI systems[15] and standards for comprehensive red-team testing to guarantee safety prior to the public release of AI systems.[16] 

II.  Advancement of Equity and Civil Rights

  • Preventing Hiring Discrimination. By October 2024, the Secretary of Labor must release guidance for federal contractors focused on preventing unlawful discrimination in hiring processes that involve AI and other technology-based hiring systems.

  • Protecting Civil Rights. In accordance with existing executive orders,[17] the Attorney General must (i) collaborate with and assist agencies in upholding existing federal laws aimed at addressing infringements on civil rights, civil liberties, and AI-related discrimination and (ii) consider offering guidance, technical support, and training to state, local, tribal, and territorial investigators and prosecutors regarding best practices for investigating and prosecuting instances of civil rights violations and AI-related discrimination, particularly in the context of automated systems.

III.  Promoting Responsible Innovation and Competition

  • The Order also emphasizes the importance of a competitive AI ecosystem and the need to reduce the risk that any single firm could gain control over key inputs (e.g., semiconductors, cloud storage, and labor). We have previously written about the top competition issues on which federal agencies are currently focused. To that end, the Order directs all covered federal agencies to develop policies and regulations that will promote competition in AI, and specifically calls on the Federal Trade Commission (“FTC”) to use its enforcement authority to prevent and/or address any unfair methods of competition. This approach aligns with an earlier executive order on promoting competition in the economy more broadly.  Taken together, the two executive orders seek to harness the collective regulatory power of the covered federal agencies to maintain a competitive balance across a number of industries.[18]

IV.  Consumer Privacy

  • The Order requires the OMB Director to mitigate AI-related privacy risks including by identifying commercially available information (“CAI”) procured by federal agencies, including CAI containing personally identifiable information in agency inventories and agency standards, and procedures related to the collection, processing, and use of CAI with personally identifiable information to develop guidance on privacy and confidentiality. The Order also requires the OMB Director to issue by April 2024 a Request for Information (“RFI”) to gather feedback on methods of improving privacy impact assessments, and to take the steps to support and advance the actions and strategies identified through the RFI process. By October 2024, the Secretary of Commerce is required to establish guidelines for agencies to evaluate the effectiveness of differential-privacy-guarantee protections,[19] particularly in the context of AI usage. These guidelines will consider key factors and common risks associated with differential privacy. Additionally, by February 2024, the Director of the National Science Foundation is required to fund a Research Coordination Network (“RCN”) to advance privacy research and privacy-enhancing technologies (“PETs”),[20] facilitating collaboration among privacy researchers and the establishment of privacy research standards, and collaborate with agencies to identify opportunities to incorporate PETs into their operations, prioritizing research efforts that encourage the adoption of advanced PETs solutions.

V.  Additional Guidelines and Principles

The Order outlines several other guidelines and principles focused on supporting workers, protecting consumers, advancing America’s leadership abroad, and ensuring the federal government’s responsible use of AI.

  • Supporting Workers. By April 2024, the Chairman of the Council of Economic Advisers must submit a report to the President on AI’s effects on the American workforce, and the Secretary of Labor is required to prepare a report analyzing how federal agencies can assist workers displaced by AI and technological advancements. This includes assessing the use of existing federal programs and potential legislative measures to strengthen support for displaced workers.

  • Consumer Protection. Independent federal agencies, including in the fields of healthcare, transportation, and education, are advised to utilize their regulatory powers to safeguard American consumers from fraud, discrimination, privacy breaches, and other potential risks associated with AI. This includes assessing the need for rulemaking, as well as reinforcing or specifying the areas in existing regulations and guidance that are applicable to AI.[21] Additionally, the Order emphasizes the obligation of regulated entities to perform appropriate due diligence and monitoring when using AI (including with respect to third-party services), highlighting existing requirements for transparency and model audits.

  • Advancing International Leadership. To strengthen U.S. leadership in the global AI landscape and effectively address its challenges, the Order requires the Secretary of State to lead efforts to engage international allies and partners in bilateral, multilateral, and multi-stakeholder forums, excluding military and intelligence domains, to enhance their understanding of existing and planned U.S. AI-related guidance and policies and foster international collaboration. Additionally, the Secretary of State will work on creating an international framework to manage AI risks and maximize benefits. This includes encouraging allies and partners to support voluntary commitments similar to those undertaken by U.S. companies and developing common regulatory and accountability principles for foreign nations.

  • Federal Government’s Use of AI. To ensure the coordinated use of AI throughout the federal government, by December 2023, the Director of OMB is required to establish an interagency council to oversee the development and deployment of AI in agency programs and operations, excluding national security systems.

[1] The Order uses the definition of AI as set forth in 15 U.S.C. 9401(3): “a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments. Artificial intelligence systems use machine- and human-based inputs to perceive real and virtual environments; abstract such perceptions into models through analysis in an automated manner; and use model inference to formulate options for information or action.” See Sec. 3(b).

[2] Companies including OpenAI, Amazon, Anthropic, Inflection, Google, Microsoft, and Meta committed to creating ways for consumers to identify AI-generated content and ensuring the security of their AI tools before their public release. These commitments include implementing methods for consumers to recognize AI-generated materials, involving independent experts to assess tool security, sharing best practices and information with industry players, allowing third-party vulnerability reporting, disclosing limitations, and prioritizing research on societal risks of AI.

[3] The Order defines “AI Systems” as any data system, software, hardware, application, tool, or utility that operates in whole or in part using AI. See Sec. 3(e).

[4] The 11 principles identified are intended to be a non-exhaustive list to guide “all AI actors” regarding the design, development, deployment, and use of AI systems. The principles are broad and address measures to address risk, mitigate vulnerabilities in systems, promote transparency through public reporting, develop governance and AI risk management policies, and implement data security controls. The voluntary code of conduct sets forth voluntary guidance for organizations developing advanced AI systems (including foundation models and generative AI systems) to promote safe, secure, and trustworthy AI. The 11 actions set forth in the voluntary code are similar to the principles and include recommendations to share information and report incidents across AI organizations, develop content authentication and watermarking procedures, and invest in mitigation efforts.

[5] In addition to the measures outlined below, the Order also includes measures to mitigate the dangers associated with the utilization of AI in engineering hazardous biological materials by establishing robust and innovative criteria for screening biological synthesis, mandates the creation of a cybersecurity initiative aimed at creating AI-based tools for the detection and rectification of vulnerabilities in essential software, and calls for the issuance of a National Security Memorandum that instructs additional measures related to AI and security to outline strategies to counter the use of AI by adversary forces for military purposes.

[6] The Order invokes the Defense Production Act, a Cold War-era law that provides the president with significant control over domestic industries. Specifically, it also allows the president to order companies to cooperate in ways that might otherwise violate antitrust laws.

[7] We note that the companies most likely to develop the next iteration of large, powerful models have all signed the July White House statement, making public and voluntary commitments to test their models before deploying them. However, this Order gives those voluntary commitments the force of law.

[8] The Order focuses on dual-use foundation models that are currently more powerful than the currently commercially available foundation models such as GPT-4 and Claude-2 (i.e., computing power exceeding 1026 floating point operations per second (“FLOPS”), which is an arbitrary threshold but nevertheless one that AI experts believe may be superseded by the next generation of large language models). The Order defines “dual-use foundation models” as an AI model that is trained on broad data; generally uses self-supervision; contains at least tens of billions of parameters; is applicable across a wide range of contexts; and that exhibits, or could be easily modified to exhibit, high levels of performance at tasks that pose a serious risk to security, national economic security, national public health or safety, or any combination of those matters, such as by (i) substantially lowering the barrier of entry for non-experts to design, synthesize, acquire, or use chemical, biological, radiological, or nuclear weapons, (ii) enabling powerful offensive cyber operations through automated vulnerability discovery and exploitation against a wide range of potential targets of cyberattacks, or (iii) permitting the evasion of human control or oversight through means of deception or obfuscation. See Sec. 3(k).

[9] The Order defines “AI red-teaming” as a structured testing effort to identify flaws and vulnerabilities in an AI system, often in a controlled environment and in collaboration with developers of AI, often performed by dedicated “red teams” that adopt adversarial methods to identify flaws and vulnerabilities, such as harmful or discriminatory outputs from an AI system, unforeseen or undesirable system behaviors, limitations, or potential risks associated with the misuse of the system. See Sec. 3(d).

[10] The Order requires the Secretary of Commerce to establish and regularly update technical conditions for models and computing clusters subject to these reporting requirements. Until these technical conditions are defined, reporting requirements will be mandatory for (i) any model that was trained using a computing power exceeding 1026 FLOPS, or using primarily biological sequence data with computing power greater than 1023 operations, and (ii) any computing cluster with physically co-located machines in a single data center, connected via data center networking exceeding 100 Gbit/s, and having a theoretical maximum computing capacity of 1020 FLOPS for AI training.

[11] The Order uses the defined term Infrastructure-as-a-Service (“IaaS”) to mean any product or service offered to a consumer, including complimentary or “trial” offerings, that provides processing, storage, networks, or other fundamental computing resources, and with which the consumer is able to deploy and run software that is not predefined, including operating systems and applications. The consumer typically does not manage or control most of the underlying hardware but has control over the operating systems, storage, and any deployed applications.

[12] These reports would be required to contain, at a minimum, information regarding the identity of the foreign party involved and the occurrence of any training run of an AI model meeting the criteria outlined in subsection 4.2(c)(i) of the Order.

[13] The Order defines “generative AI” as the class of AI models that emulate the structure and characteristics of input data in order to generate derived synthetic content, including images, videos, audio, text, and other digital content. See Sec. 3(p).

[14] The Order defines “model weight” as a numerical parameter within an AI model that helps determine the model’s outputs in response to inputs. See Sec. 3(u). Model weights are learnable parameters that are used in machine learning models such as neural networks.

[15] These include developing a companion to NIST’s AI Risk Management Framework (“NIST AI RMF”), creating a software development framework to incorporate secure development practices for generative AI and dual-use foundation models, and launching an initiative to create guidelines on evaluating auditing AI capabilities, focused on cybersecurity and biosecurity. In January, NIST released the long-awaited AI RMF, the AI governance framework resulting from a multiyear process involving industry participation and feedback. The AI RMF was designed with the goal to help increase trustworthiness of AI and to manage risks associated with the development and use of AI. The AI RMF is not legally binding or required for AI development and deployment, but has become a de facto standard for AI governance.

[16] These standards include developing guidelines assessing and managing the safety of dual-use foundation models and develop testing environments and coordination with the Secretary of Energy and the Director of the National Science Foundation (“NSF”) to develop testing environments, such as testbeds.

[17] Executive Order 12250 of November 2, 1980, Executive Order 14091 and 28 C.F.R. 0.50-51,

[19] The Order defines “differential-privacy guarantee” as protections that allow information about a group to be shared while provably limiting the improper access, use, or disclosure of personal information about particular entities. See Sec. 3(j).

[20] The Order defines “privacy-enhancing technologies” as any software or hardware solution, technical process, technique, or other technological means of mitigating privacy risks arising from data processing, including by enhancing predictability, manageability, disassociability, storage, security, and confidentiality. See Sec. 3(z).  

[21] Over the last few years, the FTC has been publishing guidance on existing regulations applicable to AI.  See, e.g., Michael Atleson, Keep Your AI Claims in Check; Federal Trade Commission Business Blog, (Feb. 27, 2023), https://www.ftc.gov/business-guidance/blog/2023/02/keep-your-ai-claims-check; Elisa Jillson, Aiming for truth, fairness, and equity in your company’s use of AI, Federal Trade Commission Business Blog, (Apr. 19, 2021), https://www.ftc.gov/news-events/blogs/business-blog/2021/04/aiming-truth-fairness-equity-your-companys-use-ai; and Andrew Smith, Using Artificial Intelligence and Algorithms, Federal Trade Commission Business Blog, (April 8, 2020), https://www.ftc.gov/news-events/blogs/business-blog/2020/04/using-artificial-intelligence-algorithms.


This communication is for general information only. It is not intended, nor should it be relied upon, as legal advice. In some jurisdictions, this may be considered attorney advertising. Please refer to the firm’s data policy page for further information.