Trump likely to scale back AI policy with repeal of Biden order

Donald Trump’s election victory means changes are coming for the government’s approach to artificial intelligence, including a promise to roll back President Joe Biden’s order outlining federal agencies’ work with the growing technology.

While it’s unclear exactly how the second Trump administration’s approach would differ, AI policy experts told FedScoop they expect something slimmer in place of Biden’s order.

Adam Thierer, a senior fellow for R Street Institute’s technology and innovation team, said it’s likely “agencies will be instructed to take a slightly more hands-off approach to AI regulation and also consider alternative approaches besides regulation,” as was the case under the first Trump administration’s own AI memorandum.

The real challenge for Trump’s team, Thierer said, is how to play a “cut-and-paste game of saying what you do and do not like as you do the repeal and replace of the executive order.”

Some actions in Biden’s order that have faced Republican criticism are likely on the chopping block, like use of the Defense Production Act and focus on algorithmic discrimination. But a Trump policy may also continue aspects of Biden’s strategy where there may be more bipartisan support. These include cybersecurity guidelines, certain national security recommendations and language about government efficiency.

The 2024 Republican party platform included plans to rescind Biden’s executive order on AI (EO 14110) calling it “dangerous” and saying that it “hinders AI Innovation, and imposes Radical Leftwing ideas on the” technology’s development. It promised to replace the order with a plan that supports “AI Development rooted in Free Speech and Human Flourishing.”

Biden’s order, released in October 2023, established the foundation of the administration’s approach to the technology. Though it sparked some regulation, the order mostly directed the creation of an internal governance structure, acquisition guidance and, most recently, a national security framework. At the time it was rolled out, the administration touted it as the strongest government action on AI taken in the world.

Divyansh Kaushik, vice president at Beacon Global Strategies who focuses on emerging technology and national security, said he expects the Trump administration to put national security front and center, which the Biden administration has also done. But the guardrails each administration imposes are likely to differ.

“The way we get there — I think that’s where you will see differences,” Kaushik said.

Areas of tension

Though neither Trump nor the Republican party have stated specifics for the coming administration’s approach to AI, conservative rhetoric around the technology points to several pain points ripe for change.

In particular, the order’s use of the Defense Production Act to require companies that develop foundation models that are risky to national security to notify the government when training the model and share the results of testing has been the subject of criticism by Republican lawmakers.

Use of the DPA has already been called out by Sen. Ted Cruz, R-Texas, and former Texas Republican Sen. Phil Gramm as a misuse of that statute that creates “frightening precedent.” Cruz also criticized the Biden order for hindering AI development and introduced an amendment in committee that would have repealed the order in its entirety. He’s a potentially important voice as the current ranking member of the Committee on Commerce, Science, and Transportation and is in line to lead the panel with the power change in the Senate. 

Matt Mittelsteadt, a research fellow at George Mason University’s Mercatus Center focused on AI, said that while the Trump administration would likely pursue deregulation of the technology, the Biden order is a largely non-regulatory document. But even removing the deployment of the DPA, Mittelsteadt said, may not do much to impact how the government approaches AI. 

There’s also the question of countering AI bias. Although the Trump White House also flagged bias and discrimination as potential issues with AI that agencies should consider, it’s likely that the coming Trump administration wouldn’t go as far as Biden’s order and that civil rights would not be front-and-center in the AI conversation. For instance, in a September letter to RAND Corp. seeking information about its relationship to the Biden administration’s AI policy and order, Cruz criticized safety standards from NIST, calling them “woke.” 

The emphasis in the order and in the AI Bill of Rights around bias and algorithms is likely to go away, cautioned Neil Chilson, head of AI policy at the emerging technologies nonprofit Abundance Institute and former chief technologist at the Federal Trade Commission. “Republicans are very skeptical of that type of approach,” he said.

Notably, a fundamental element of the Office of Management and Budget’s corresponding guidance (M-24-10) to Biden’s executive order is a risk management structure based on whether uses might impact Americans’ rights or safety. If a use falls under one of those categories, agencies must ensure they meet certain requirements, such as ensuring those uses mitigate algorithmic discrimination and advance equity. That framework continued into OMB’s recent guidance on how the government should buy AI. 

A general criticism of those requirements is “that they’re creating more red tape for the agencies to use AI,” Kaushik said. “I think that’s one of the worries the incoming administration will share.”

The talent portion of the order that leaned into attracting workers from foreign countries could also be an area where the Trump administration might differ, he added.  

AI Safety Institute, NAIRR future

A full rescission of the Biden order could also affect actions that have bipartisan support, such as the Department of Commerce’s AI Safety Institute and the National AI Research Resource pilot, which is being run by the National Science Foundation. 

The AISI, housed in the National Institute of Standards and Technology, is focused on evaluation, testing and developing guidelines for AI models and has been central to Commerce’s actions on the technology. The NAIRR, meanwhile, is a resource aimed at providing tools needed to advance AI research. What the incoming administration does with those activities will largely depend on Congress’ appetite to advance bipartisan legislation to codify the institute and research resource. 

The pending AI Advancement and Reliability Act (H.R. 9497) and Future of Artificial Intelligence Innovation Act (S.4178) would authorize the AI Safety Institute and have been favorably reported by committees in each chamber — though the House bill would do so under a different name. The CREATE AI Act (H.R. 5077; S. 2714) that would codify the NAIRR has also advanced out of committee in the House and Senate.

With respect to the AI Safety Institute legislation, Kaushik said “it’s a bit hard to see that the authorization actually happens this Congress.” The NAIRR legislation could advance, he said, but what is more likely is that must-pass legislation at the end of the year remains “clean,” particularly under Republican control of Congress.

Thierer said Trump may want to pull back some of the authority of the AI Safety Institute and return it to the Office of Science and Technology Policy, which was where AI decisions were centralized during the previous Trump administration. “We might see a return to that model of White House leadership, as opposed to delegating it to the Department of Commerce,” he said.

Trump’s relationship with Elon Musk could also play a role, said Stephanie Haven, a UC Berkeley tech policy fellow who is focused on the impact of generative AI in war and conflicts.

As long as Trump and Musk are connected, the SpaceX and Tesla founder’s own approach to AI safety “will play a significant role in shaping Trump’s approach,” Haven said. As a result, the voluntary commitments the institute has secured around testing could stick.

Similarly, the NAIRR has quite a bit of bipartisan support, and the pilot is the product of the recommendations from a task force that was also created under the  National Artificial Intelligence Initiative Act, which was signed into law by Trump. The NAIRR is often described as a shared research infrastructure with the goal of democratizing access to things like computing power and data that are needed for AI research. 

As a pilot project, however, it needs more support to continue. The Biden administration has repeatedly called on Congress to further support the initiative so it can continue its work. 

Mittelsteadt said codifying the NAIRR is very likely to happen at some point given its extremely bipartisan support; it’s just a matter of when Congress acts. Even in the event that the EO goes away and it’s briefly disestablished, Congress is likely to put it back together. “I would say it’s by far the most bipartisan piece of the puzzle here,” he said.

The incoming administration may also want to reframe the NAIRR. Chilson said right now the resource is framed as something that could advance research on things like algorithmic bias. “The Trump administration will have very little interest in that,” Chilson said. What they might have an interest in is how to use the resource to ensure the U.S. stays ahead of China in AI technical and capability development, he said

National Security, governance

While national security is likely to be a focus for Trump on AI — particularly as the U.S. races to compete with China — it’s not yet clear whether the incoming administration will also want to roll back Biden’s efforts on that topic.

The Biden administration recently released its national security memorandum on AI, creating a roadmap for national security strategy on the technology with a focus on the intelligence community and Defense Department. That memo, which was a requirement of the AI order, also created a central role for the AI Safety Institute.

Mittelsteadt said he hopes “that just because they oppose the EO, that doesn’t necessarily mean they’re going to just throw out the playbook on national security.” But he added that whatever is in the memo that depends on the order might have to be revised if it’s rescinded.

Kaushik noted that discussions about competition with China and use of the technology in that context is “the Biden administration continuing the policies of the Trump administration.” It’s also very rare for administrations to revoke an NSM from a previous administration, he said.

“The Biden administration has done a pretty good job of future-proofing the NSM because they’ve kept the NSM to things that are generally bipartisan,” Kaushik said. Meanwhile, he said the guardrails for national security and prohibited uses were kept separate from the memo in a corresponding framework. 

One space to watch in that area will be open source, according to Thierer, who noted there seems to be “countervailing forces at work in the Trump administration behind the scenes” on the topic. Vice President-elect JD Vance has been publicly supportive of open-source AI — or models with code publicly available — as a counter to big tech. But at the same time, those in the national security community and defense space see open source as a potential vulnerability.

Some of the governance efforts in Biden’s order and the OMB memo are similarly expansions of Trump administration approaches. Trump’s own December 2020 order on government use of the technology (EO 13960), for example, established some principles for responsible use, transparency, and having humans in the loop that Biden expanded upon.  

The public inventory process for AI use cases within agencies was established in Trump’s order on the technology and was later expanded and refined by the Biden administration. And the work of the responsible AI officials created under the Trump order were, in many agencies, absorbed into the responsibilities of the chief AI officers required by Biden’s order. 

“This is an area in which I expect both a certain degree of continuity and a certain degree of change,” Mittelsteadt said of the Biden OMB guidance. He said the Trump administration will likely wipe the guidance for those uses clean and its approach might be “less onerous” or even delegated to the agencies to figure out.

Kaushik noted that many of the Biden administration’s actions on federal AI use are still in early phases. Agencies, for example, just released plans for compliance with the memo and the deadline for the annual inventory is in December. 

Because of that, he said, it might be an area where there isn’t much difference.

FedScoop reporters Caroline Nihill and Rebecca Heilweil contributed to this article.