Powerful Cold War-era tool Trump could use in an AI emergency is set to expire in December

Trump Administration Eyes Defense Production Act to Tame Emerging AI Threats

WASHINGTON — In a bid to keep rapid advances in artificial intelligence (AI) from becoming a national‑security hazard, the Trump administration is turning to one of the country’s oldest weapons‑protection tools: the Defense Production Act (DPA). The legislation, originally written in 1950 to marshal U.S. industry for the Korean War, has long been used to ramp up military production, prioritize federal contracts and respond to emergencies. Experts say its broad authority could be leveraged to regulate the fastest‑growing frontier of technology.

The DPA’s powers extend far beyond weapons manufacturing. They include provisions for dealing with cyberattacks, controlling exports of sensitive technology, imposing conditions on federal contracts and imposing civil liability. “The authorities could be a little clearer,” said James Lewis, director of the Strategic Technologies Program at the Center for Strategic and International Studies. “But you could probably cobble together a solution using the legal authorities we have now.”

The urgency of using the DPA came after a group of CEOs from some of the country’s most influential AI firms issued warning statements this week about models that are becoming increasingly capable. They urged regulators to slow the pace of development to give safety measures time to catch up. Trump dismissed the calls for new guardrails, stating the administration already has “tremendous criminal and regulatory power” over AI companies.

The DPA was due to expire on September 30, 2025, but Congress extended most of its authorities until December 11, 2026. The extension is a temporary measure that will likely be contested later this year, just before the political landscape could shift following the November midterm elections.

A Framework for Early Detection

The administration’s latest move was a June executive order that directs federal agencies—including the National Security Agency (NSA) and the Cybersecurity and Infrastructure Security Agency (CISA)—to develop classified benchmarks for advanced cyber capabilities. The order also defines what constitutes a “covered frontier model,” a category of AI systems deemed especially dangerous. The order does not mandate pre‑approval or licensing but encourages developers to voluntarily provide the government with early access to such models.

“We’re trying to establish a system where the government can see the front‑line of technology before it can cause harm,” said a senior official from the Department of Homeland Security. “It’s a voluntary framework that gives developers the chance to cooperate before releasing a potentially risky model.”

But identifying a dangerous capability is only the first hurdle. The next step is deciding what the government can actually do. Legal scholars point to the DPA’s information‑gathering authority as a key tool. James E. Baker, former chief judge of the U.S. Court of Appeals for the Armed Forces, said the act could compel companies to report what they are building: “What you could use the DPA for is to require reporting, and that's the first step to making sound policy. You can have a debate about whether there should be regulation and what the regulation should look like. But you really can't argue against knowing what's happening.”

Limits and Potential

Paul Rosenzweig, a lawyer and former deputy assistant secretary for policy at the Department of Homeland Security, cautions that the DPA was designed primarily to ensure the U.S. has the goods and services it needs for defense, not to stifle private innovation. “The Defense Production Act is more about mandating the production of things for the United States rather than preventing the production for the general public,” he explained. “It does give them a pretty good handle for demanding more information about a defense‑sensitive kind of product, which frontier AI certainly could be reasonably characterized as.”

Once the government knows what an AI company is developing, a range of other authorities can be invoked. Export controls can prevent sensitive technology from falling into the hands of adversaries; federal contracting rules can impose conditions on companies seeking government business; and criminal law can be applied to the use of AI in illicit acts. The June order also directed the attorney general to prioritize enforcement of existing federal criminal statutes against individuals who use AI to illegally access or damage computer systems.

In addition to criminal law, federal and state governments possess a patchwork of legal tools that could apply to frontier AI—tort liability, privacy regulations, consumer protection statutes, and more. “There is no broad AI exemption from generally applicable law,” Rosenzweig said. “But no one knows precisely how well that patchwork would hold up in a true AI emergency because it has never been tested.”

The Challenge of Autonomy

A key concern is what happens when an AI system acts on its own accord and causes damage. If an autonomous model hacked a hospital, company, or critical infrastructure network, courts would have to decide whether the developer was negligent, whether a stricter product‑liability standard should apply, and whether the damage could be traced back to a particular AI system. “I tend to think that liability is going to be difficult to establish early on,” Rosenzweig added. “Right now, that's not where I would base my deterrence model.”

The uncertainty underscores the limitations of the existing toolkit. “The laws already on the books give federal and state authorities multiple avenues to respond to dangerous uses of AI, but many have never been tested against a system acting with increasing autonomy,” said Rosenzweig. “We’re applying laws written for other threats to a technology that is evolving faster than the legal system around it.”

Looking Ahead

As the Trump administration pushes forward with its plan to harness the DPA, AI industry leaders and policymakers will watch closely. The extended DPA, the June executive order, and the broader set of legal authorities will determine how the United States balances the promise of AI innovation with the imperative to guard national security. Whether the Defense Production Act can be adapted to regulate an intangible, rapidly evolving technology remains to be seen. The next few months—and potentially the 2026 reauthorization debate—will be critical in