Opportunities and risks inherent in Trump's AI Action Plan

President Donald Trump released a new AI (AI) action plan on Wednesday, coinciding with the “Winning in the AI Race” summit in Washington, D.C. The 28-page document includes more than 90 policy recommendations that believe it will expand global sales of US AI technology, speed up data center construction, and reduce “red tapes,” which have proven obstacles to the AI industry.

As Trump put it in his keynote speech, the plan will help the US “win with AI while dismantling regulatory barriers.”

But can the administration encourage rapid AI scaling both inside and outside the government without new funding, and plan more risk assessments, bad actors and other issues? Some experts also believe that other White House policies could derail some of Trump's proposed AI efforts.

Following the summit and the president's speech, the CFR convened seven experts to examine the plan of action and detail the opportunities and risks to foresee.

Trump's plan can open AI Pandora's box

Sebastian Marabie – Paul A. Volker Senior Fellow, International Economics

Under President Biden, the US government's AI policy balances three objectives. It promotes rapid adoption of AI across the economy, thereby boosting growth. Mitigates risks from both the abuse of AI by bad actors and the benefits of humans and the evil AI systems. Under President Trump, the first two objectives are promoted at the expense of a third objective.

The clearest indication coming out of the Trump administration's new executive order is its stance on open source and basic models of open weight. These types of models can be freely downloaded and adapted by anyone, making them popular with academics, entrepreneurs, and others who want to build AI applications cheaply. The downside is that open models (although they promote the spread of AI through the economy) also increase risk. Bad actors may use them on the bad edge, from mass production of deep fakes to construction of weapons.

The Biden administration also shunned the challenge of closing the open model by suggesting that future administrations may have to reconsider the questions. The Trump administration has clearly decided to stick to openness. It accepts extra risk in exchange for its faster adoption of AI and stronger capabilities to compete with China. This also uses an open model.

An important first step to suppose important questions

Jessica Brandt – Senior Fellow in Technology and National Security

The AI Action Plan correctly emphasizes the importance of enabling the private sector to proactively defend AI innovation against security risks, particularly from the threats of malicious cyber actors and insiders. If Beijing or other competitors can simply steal the latest trained models and enjoy the benefits of US investments while avoiding restrictions, investing billions of dollars in AI infrastructure and tightening export controls with sophisticated chips would be of little help. The plan provides little details on how management intends to advance its goals, but there are many measures to consider, including robust threat intelligence sharing mechanisms and incentives for secure development.

The plan also brilliantly highlights the importance of assessing national security risks from the frontier model. This is an essential step to alleviating them. However, if the assessment finds that the model exceeds the ability or risk threshold, what will happen remains unknown, suggesting that it can cause significant actual harm. Under their safety framework, major frontier companies are committed to suspending deployments that do not have sufficient mitigation, but there is no common understanding of what those mitigations should involve or who determine that is sufficient. That issue can quickly become an urgent one for the national security policy community, so it's important to start working on it now.

Is this AI action plan sustainable?

Michael C. Horowitz – Senior Fellow in Defense Technology and Innovation

The Trump AI Action Plan is a story of two impulses within the administration. The majority of action plans, including many of the pilot projects focused on open source and open weight models, are positive developments that promote employee AI literacy, assess national security risks of frontier models, and are consistent with the ongoing two-tier approach to AI's US leadership. However, there is also the risk that many of the actions requested, such as cutting technological research infrastructure within and outside the government, are either lacking certain funding or lacking potential macro tensions with other Trump administration priorities, leading to no trace.

For the Department of Defense, the most important recommendation is to create a virtual certification ground for AI and autonomous systems. This is essential for the type of AI system testing and evaluation required to create reliable reliability and scale adoption. With that and other recommendations about the Pentagon, it is important to check if you are following your resource and organizational prioritization.

The document also reflects the ongoing tensions that the administration is still striving for American AI policies overseas. For example, the administration has the goal of combating China's influence in international organizations that influence AI conferences. However, this requires the administration to provide funding and do its job within international organizations such as the United Nations agencies and the International Telecommunication Union, and within international organizations that are opposed to its policies.

Trump's AI plan begins to progress and spits diplomacy

Kat Duffy – Senior Fellow of Digital and Cyberspace Policy

A “winning” AI race requires rapid, coordinated support for AI innovation, infrastructure and international engagement. The AI Action Plan establishes important and constructive priorities by committing to improving domestic AI production capabilities, supporting systematic assessments of at least some of the risks of frontier models, expanding the scaling of American national artificial intelligence research resources, speeding up regional AI centers for outstanding model testing, and expanding access to computing power by preparing American workers for the upcoming technology transition.

It also proposes federal government incentives to improve and promote the provision of “full stack” AI in the United States. While this could prove a controversial proposal, it positions American AI companies to compete more effectively against China's highly subsidized products.

Unfortunately, the international strategy of AI action plans is undermined by dogmatic rhetoric and competing impulses. Even longtime allies and partners have been framed as mere markets to be arrested for global American hegemony. “Leberagg[ing] The US position in international diplomacy and standard setting agencies” is complicated by the recent State Department shootings disproportionately targeting individuals and offices with technical expertise, and the US decision to withdraw from major multilateral agencies such as the World Health Organization and UNESCO. The plan is trying to complicate Washington's “Global Alliance.”

Ultimately, when it comes to global engagement, the plan sings a zero-sum tune. Such framing may resonate domestically, but it will become incongruous overseas.

AI could provide national security benefits and could do great harm

Erin D. Dunbacher – Stanton Senior Fellow of Nuclear Security

It is true that AI could transform combat and back office operations across the US military. If AI can make processes and information sharing more effective, experts need to test, train and implement. With weapons, faster data processing and delegation of tasks to AI can build resilience. A more undeniable system could diversify options for leaders amid the crisis.

However, there are areas where aggressive AI adoption is not useful for national defense or security interests. AI off-limits should be where strategic early warnings, qualitative intelligence analytics capabilities, and where the challenges of “black box” and vulnerabilities inherent in all AI can be insurmountable or catastrophic. Department of Defense AI needs to make Americans safer.

Finally, in order to preempt the way AI drives the development of weapons of mass destruction (chemicals, biology, radiation, or nuclear), the administration must evaluate systems beyond partnerships with frontier AI companies. Policy should begin by limiting the way that the Language Learning Model (LLM) is most likely to share dangerous or harmful information.

Trump's AI action plan is in conflict with Trump's policy

Rush Dosi-CV Star Senior Fellow of Asian Studies and Director of China Strategic Initiative

The AI Action Plan begins by declaring that “America depends on the competition to achieve global domination in AI.” Deny access to enemy access, reject stronger export controls for semiconductor manufacturing equipment, and steps to ensure that advanced chips do not end in China.

However, the plan is strained with policy. The Trump administration overturned China's ban on selling sophisticated American chips (Nvidia's “H20”) and promoted the Chinese model. Supporters of the reversal say that only ongoing sales will prevent China from replacing Nvidia chips with Huawei's. Opponents say the risks are exaggerated, and the best way to ensure that Huawei chips are not sufficient and rely on American technology is to export the entire AI stack (such as US chips, US cloud, US data centers) and not something the enemy can use to build their own stacks.

Regardless of where the debate is, the question now is whether this decision is a sign of a more tolerant policy or alternatively a singular action. Despite Page 30 and the accompanying public comments by President Trump and Vice President JD Vance, the new AI Action Plan has yet to resolve that question.

Shot across a Chinese digital bow

Jonathan E. Hillman – Senior Fellow in Geographical Economics

The AI Action Plan clearly aims to China's Digital Silk Road, a signature initiative of China's leader Xi Jinping. One of the more promising international recommendations in the plan is the proposal to establish a program to export the US full AI technology stack (hardware, software, training and most importantly fundraising) to overseas partners. Despite security concerns, Chinese technology providers, especially Huawei, have continued to be successful in foreign markets due to their ability to provide the complete package.

This part of Trump's AI plan could be even stronger if it streamlines interagency processes and focuses more narrowly on emerging markets. The long list of US institutions mentioned in the recommendations can otherwise be slow to implement and create challenging ideas. Increased geographical focus can also be helpful.

It is emerging markets where Chinese tech companies are making the deepest invasions, particularly in the small, developing economy where China's choices can be the only one on the table. Expanding the availability of affordable US technology in emerging markets would be a major victory for the US both commercially and strategically.