On October 30, 2023, literally on the eve of the world’s first AI Safety Summit in the U.K., President Biden signed a sweeping executive order on Artificial Intelligence (AI). By doing so, the United States is attempting to position itself as a leader in the race to regulate AI. The Executive Order seeks to strike a balance between incentivizing innovation of this fast-developing technology and guarding against its numerous dangers. Executive Orders can be reversed so it remains to be seen whether this one leads to congressional regulation. While generally lacking in concrete requirements for the public, the Executive Order outlines the administration’s thinking on key issues arising from generative AI and how it might begin to regulate this world-changing technology. Some of the key takeaways are summarized below.
Directs the creation of new watermarking standards for generative AI
The Executive Order attempts to reduce AI-generated misinformation by creating national standards for identifying content created by AI. This “watermarking” has been discussed since generative AI was introduced, but there is still no reliable way to detect whether a particular piece of content is AI- generated. This allows for AI to be used as a tool to assist in massive misinformation campaigns, deep fakes, scams, false advertising, and numerous other bad acts. Presumably, these threats could be fixed by clear and consistent methods of marking AI-generated content.
This order takes the first step at creating uniform and trusted systems for determining the authenticity of content. While the order stops short of creating a mandate on watermarking, it directs the Commerce Department to draft guidance for labeling and detecting AI-generated material. This guidance will, hopefully, help foster industry standards for determining the authenticity of content. Microsoft and Google have already taken steps at creating watermarks for AI-generated content, but the order should broaden acceptance of the technique.
Places requirements on owners of large language models
To address national security issues (e.g., using AI to facilitate hacking, creation of biological weapons, etc.), one of the only firm mandates in the Executive Order is the requirement that owners of large language models of a certain size must share their safety testing results prior to releasing the technology to the public. Specifically, the order requires, “on an ongoing basis,” reports on training, developing, and producing generative AI, results of “red team” testing, and model weights. While these requirements do not seem onerous, companies who have or plan on acquiring a large language model should take note.
Directs agencies for guidance on IP-related issues
The President has also taken note of the panoply of IP-related issues with AI, a topic we have written about previously. Specifically, the Executive Order requests guidance from the United States Patent and Trademark Office regarding inventorship and the use of AI. As we previously noted, AI-generated content is not eligible for patent protection, but the level of AI assistance allowable in the patent process is an open question. The Executive Order suggests this administration wants some clarity on that question.
The Executive Order also addresses the concern of IP theft resulting from generative AI. The order instructs the Secretary of Homeland Security to develop a task force for “combatting AI-related IP risks.” While again short on details, this taskforce will certainly have to address the flood of IP-theft that will come from generative AI drastically reducing the time and effort required to steal another’s trademark, likeness, copyrighted work, etc.
Calls for privacy legislation
Finally, although not in the text of the Executive Order, the President called on Congress to pass “bipartisan data privacy legislation.” We previously wrote about the bipartisan draft No Fakes Act. That draft bill in essence creates a national right of publicity, which greatly expands on the various states’ laws. The President unsurprisingly appears to support that bill, and presumably wants broader privacy-based legislation as well.In total the Executive Order appears to be a good first step in combatting the various concerns with AI, while also not stifling its development. But the devil is in the details, and the details are not here. But it’s a helpful look at where this is headed.
Sign up for our newsletter and get the latest to your inbox.