Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence
In the days preceding the first global summit on internet security (a subject for imminent blog commentary), and in a timely strategic move, the White House published the first comprehensive federal instrument on AI. I have purposely avoided the term ‘norm’ because an Executive Order does not exactly equate to a regulatory decree, let alone a Spanish decree-law or legislative act. It principally sets forth principles and, most importantly, instructs the Administration to undertake various actions.
However, this Executive Order is much more. It seeks not merely to act as a starting point for subsequent legislative action (a ‘kick forward’), but on one hand, it contains a declaration of objectives so thoroughly developed and solid that it approaches a strategic and national legislative program; on the other hand, it delves into conceptual and technical details that will likely permanently imbue the potential bipartisan debate; and finally, it establishes specific bodies, mandates the issuance of reports, and guides the actions of various powerful administrative agencies, which possess and typically exercise their competencies with vigor.
This is evident on the explicit or manifest level. However, perhaps the main strength of this Executive Order lies in its implicit, understood value: it is unlikely that any executive or legislative action will escape its hermeneutic scope, particularly in the case of legal gaps (predictable in a technology as dynamic as AI, and therein lies the problem).
As has been announced, it is an all-encompassing instrument: only the previous project of the AI Rights Charter and certain prior sector-specific norms are excluded. In fact, as will be seen, this prior legislation is somewhat subsumed within EO 14110.
I do not intend to analyze the Order in detail; you can find more substantive summaries and commentary in the sources listed in the bibliography. However, I do wish to highlight, in general terms, its most notable aspects.
Firstly, the Order establishes eight guiding principles on AI that are to direct the Administration’s actions. These principles, not detailed here due to space constraints, range from the safety and reliability of AI systems to the protection of civil rights and privacy, fostering innovation and competition, supporting American workers, promoting equity and civil rights, protecting the interests of American consumers, managing risks in governmental use of AI, and the U.S.’s global leadership.
Subsequently, the EO includes a glossary worth reading as it contains the best theoretical and scientific practices on AI with a regulatory focus. For those uncertain about what constitutes a ‘dual-use model’ or ‘generative AI,’ this is the place to consult.
Regarding the crucial element of security, the USA appears to be moving towards a quadruple control procedure: establishment of industry standards, self-regulation, auditing, and executive intervention by the Administration (through Agencies). Integrated (‘internal’) security systems have not yet been designed – even if they were conceivable -, but in exchange, demands are disseminated – that is, multiplied – throughout the Public Administration.
A general duty of information is established for companies that are heading towards, even in the project phase, dual-use models or, regardless of use, surpass certain AI power thresholds. The risks of discrimination or bias, privacy, attacks on strategic structures, and particularly the combination of AI with biotechnology, are subject to special caution.
Clearly, the general obligations of the private sector will be nuanced in cases of national security (military and intelligence); moreover, the drafting of a National Security Memorandum is expressly contemplated to guide the use of AI in military and intelligence missions, with some precepts of guarantees noting an exception for ‘national security.’ This will be a slippery territory.
In general, it is evident that the USA has decided that this time, it will be the one, and not, for example, the EU (as in data protection) or China, to set global standards, in line with its evident technological leadership in AI.
Las Palmas, November 10, 2023
BIBLIOGRAPHY
Suleyman, M. (2023). The Coming Wave: Technology, Power, and the Twenty-first Century’s Greatest Dilemma. Crown.”
