Tech leaders urge a national AI data reserve, NIST standards and red‑teaming by national labs to keep AI development in the U.S.
Loading...
Summary
Founders and researchers at a House hearing urged Congress to create a national AI data reserve, resource NIST for measurement science, and use national labs for red‑teaming sensitive models. They said clear federal data and testing standards would help preserve U.S. leadership and democratic values.
Industry and research witnesses told the House Energy and Commerce Committee that American AI leadership depends as much on access to data, measurement standards and safety institutions as it does on chips and energy.
Alexander Wang, founder and CEO of Scale AI, urged Congress to stand up a national AI data reserve to make government datasets ‘‘AI ready’’ and more widely available to advance U.S. models. Wang said the U.S. government is itself one of the largest producers of data and that centralizing and preparing those datasets would strengthen U.S. machine‑learning work.
Wang and other witnesses advocated three linked steps: - Resource NIST and the global network of AI safety institutes to produce measurement science and test regimes the world can adopt. - Require and fund red‑teaming and third‑party safety evaluation, including classified work where needed with national labs and subject‑matter experts (for example nuclear and biological safety specialists). - Pursue a single federal AI governance standard to avoid a patchwork of state laws, while allowing sector‑specific rules for high‑risk uses in health care, financial services and critical infrastructure.
Why it matters: Witnesses argued that China’s decade‑long investments in public datasets give it a structural advantage on the raw inputs that power AI models. Alexander Wang told the committee that China is ‘‘out investing us in data’’ and that a coordinated U.S. approach to making government data AI ready and sharing standards internationally could blunt that edge.
Security and safety Several witnesses emphasized that private companies alone do not have all requisite expertise to fully vet models for national security risks. David Turk and others urged leveraging national labs and government experts for red‑teaming before large models are released publicly. Schmidt and others raised concerns about open‑source releases and techniques such as distillation that could allow foreign models to reproduce capabilities trained on U.S. data.
Privacy and consumer protection Committee members and witnesses also emphasized the need for national privacy rules so Americans' data is protected and to avoid a patchwork of state laws. Witnesses supported data minimization principles for consumer protection while preserving designated government datasets for public interest research and national security purposes.
Next steps Witnesses recommended specific congressional investment in NIST and national labs to accelerate measurement science and testing, and asked Congress to consider legislation to create or authorize a national AI data reserve and to preempt a proliferation of inconsistent state standards.
