AI, Encryption, and Data Flows: Policy Imperatives for U.S. Leadership
Editor’s note: In February, The National Interest organized a symposium on the U.S.-China technology race amidst the emergence of DeepSeek and ongoing legal battles over TikTok. We asked a variety of experts the following question: “What are the three most important technology policies that the U.S. should pursue or avoid to compete adequately with China?“ The following article is one of their responses.
Clarify and Expand Access to Training Data for AI Models
For policy related to the three pillars of artificial intelligence (AI) development (algorithms, compute, data), data access is highly sensitive to changes in the legal landscape. To this point, cutting-edge AI models have required millions to billions of pieces of diverse data over the course of the training process. In the US, leading model developers have largely relied on a mix of publicly available data scraped from the internet, as well as proprietary data acquired through licensing deals, and existing first-party data caches. The problem, however, is that every leading US model developer is being sued for copyright infringement for using data scraped from the internet.
If plaintiffs in these lawsuits prevail, the fines and other legal action could destroy much of the progress American AI model developers have made, as well as impede future AI development in the US. As federal courts wrestle with these cases, Congress should create an exception within existing copyright laws, a form of text and data mining exception aimed specifically at information used to train AI models, while also supporting the creation of technical standards to give individuals the ability to opt out of having their information used in model training. China, as well as some US allies, have updated their laws to promote model development domestically by clarifying how existing intellectual property (IP) law coexists with using data to train AI models. Falling behind the rest of the world, including our strongest competitor, when it comes to data access, could do irreparable harm to America’s future AI ambitions. Remedying this problem and balancing access to information and protections for IP in the age of AI is critical for US technological competitiveness.
Avoid Regulations Undermining End-to-End Encryption
Cybersecurity is an increasingly important paradigm for lawmakers interested in protecting the American public as well as competing in geopolitical competition with China. End-to-End Encryption (E2EE), a type of messaging that ensures information sent between two parties is private from everyone but the two exchanging parties, is vital to protecting non-public information and ensuring individual privacy. A rash of cyberespionage operations conducted in recent years by state-sponsored hacking groups aligned with nations such as China, Iran, and Russia, as well as independent groups interested in extorting public and private actors, have put millions of Americans’ personal information at risk. Such operations have also created vulnerabilities in critical infrastructure around the country, leading the US government to recommend private encrypted messaging services for official business.
At the same time, legislative efforts in the US and abroad, most recently in the United Kingdom, have attempted to force software providers to create “backdoors,” for law enforcement to access encrypted networks to prevent the use of such tools for illegal means. While these proposals are looking to solve a serious problem, the second-order effects of destroying E2EE would be significant. This would make millions of Americans and people around the world more vulnerable to cyber attacks. Undermining digital tools that can be trusted to support American digital security, anonymity, and privacy would be a boon to our Chinese competitors, not to mention other nations that would benefit from American companies’ and government actors’ weaker information security. Domestically, state and federal lawmakers should not advance legislation that would undermine services that utilize E2EE. Globally, the United States should dissuade foreign governments from passing laws that would require American firms to create backdoors into encrypted services.
Strengthen Agreements Related to Cross-Border Data Flows
Access to high-quality training data is an imperative for continued development of AI models for private and public-sector use. This is true for the initial pre-training process, where a large, unstructured data set is used for representation learning, as well as later steps such as fine-tuning, where a specialized data set is used to adapt the model to specific tasks, or validation, where a new, unseen data set is used to test the model and prevent over-fitting (a model memorizes data but is not able to extrapolate and perform generally). One way the US government can support the development of AI models is by creating bilateral and multilateral agreements focused on procuring datasets and promoting cross-border data exchanges for the express purpose of training AI models.
Such agreements would promote US competitiveness with China in two ways. First, they would respond to Chinese policies that are seeking to empower their own AI developers with data from the state. Local governments in China have been directed by the Ministry of Industry and Information Technology to make specific datasets available to model developers in areas related to healthcare, finance, labor force, and industrial operations. By working with foreign partners to create such agreements, US firms would have access to more robust and diverse pools of data for model training and fine-tuning for sector-specific uses. Second, enacting such agreements would build opportunities for collaboration with allies on AI model development. This could help open up new markets for US companies as well as avenues for further research and development for public actors and researchers. By building upon existing CBDF agreements and identifying new opportunities, the US can strengthen its relationships with nations around the world, promoting commercial and strategic partnerships related to digital commerce and governance.
About the Author: Joshua Levine
Joshua Levine is a Research Fellow of Technology Policy at the Foundation for American Innovation.
Image: Shutterstock