The Defense Production Act of 1950 (DPA; 50 U.S.C. §4501 et seq.) is a Korean War-era US statute that grants the President broad authority to shape domestic industrial activity in the interest of national defense. It is a standing emergency industrial-mobilization authority, and it has become a recurring instrument in AI and semiconductor policy.
Status and scope
The Act is not permanent; Congress periodically reauthorizes it. The term "national defense" has been construed expansively over time, to include emergency preparedness, critical-infrastructure protection, and, increasingly, technology-supply-chain security. That expansive reading is what allows the DPA to reach AI labs, though the reach is contested.
Key authorities
The DPA's provisions are organized into titles, three of which are most relevant:
- Title I — Priorities and Allocations. Lets the government require companies to prioritize and accept contracts for goods and services deemed necessary for national defense, and to allocate materials and facilities. It is also the title under which the government can compel information from industry in support of defense-related programs. Title I reporting is mandatory, unlike the voluntary frameworks (Responsible Scaling Policy (RSP), summit commitments) that otherwise dominate US practice.
- Title III — Expansion of Productive Capacity and Supply. Authorizes financial incentives, including loans, loan guarantees, direct purchases, and purchase commitments, to build or expand domestic production of critical goods. It has been used for defense-industrial-base investments, including in advanced manufacturing.
- Title VII — General Provisions. Includes voluntary-agreement authority, the small-business provisions, and definitions; historically associated with industry-coordination mechanisms.
Relevance to AI policy
The DPA's central appearance in AI governance is Executive Order 14110 (RESCINDED) (Biden, October 2023). That order invoked DPA Title I authority to require companies developing the most capable dual-use foundation models, and companies operating large compute clusters, to report to the Department of Commerce: the existence and details of large training runs, the physical and cybersecurity protections on model weights, and the results of red-team safety testing. This was the first time the federal government used a compulsory legal authority, rather than voluntary commitments, to obtain visibility into frontier-model development.
That requirement's status changed with the politics. Executive Order 14179 — Removing Barriers to American Leadership in Artificial Intelligence (Trump, January 2025) revoked EO 14110, removing the DPA-based reporting mandate. The episode has been cited as an illustration of how much of US frontier-AI oversight rested on executive action over a borrowed statutory authority rather than on AI-specific legislation, and therefore how readily it could be undone. Using a 1950 defense-mobilization statute to govern AI makes oversight contingent on executive will and vulnerable to reversal.
The DPA is also relevant to AI through its use in semiconductor and critical-mineral supply chains and domestic-capacity programs, which bear on Compute Governance and the broader AI and National Security picture.
Relationships
- invoked-by: Executive Order 14110 (RESCINDED) — used DPA Title I to compel frontier-model reporting
- superseded-in-effect-by: Executive Order 14179 — Removing Barriers to American Leadership in Artificial Intelligence — revoked EO 14110 and the DPA-based mandate
- related: Compute Governance, AI and National Security, AI Governance (umbrella)