Trump’s ‘Woke AI’ Efforts Should Focus on the Real Problem
On July 23, the Trump administration released a major artificial intelligence (AI) policy blueprint titled “Winning the Race: America’s AI Action Plan.” This plan was the culmination of a major proceeding launched in February that generated over 10,000 public responses. As noted in a new R Street analysis, the AI Action Plan establishes a light-touch AI policy agenda to help unleash AI competition and investment while enhancing national strength. The Action Plan extends the market-oriented policy blueprint the Trump administration has sketched out since retaking the White House, and additional steps are planned to build out this policy agenda.
However, one of those steps raises concerns about government overreach and could counter much of the good the administration hopes to achieve with its AI Action Plan. President Donald J. Trump signed a new executive order (EO) titled “Preventing Woke AI in the Federal Government.” The AI Action Plan recommends eliminating the terms “misinformation, Diversity, Equity, and Inclusion [DEI], and climate change” from the National Institute of Standards and Technology (NIST) AI Risk Management Framework (AI RMF). It also recommends that any frontier large language model (LLM) company contracting with the federal government be “free from top-down ideological bias” and comply with the “Unbiased AI Principles” of “truth-seeking” and “ideological neutrality.”
Neutrality and fairness in technological systems are worthy aspirational goals, but they do not translate into concrete, easily enforced public policies. Government efforts to police “fair” speech in AI systems—even through procurement processes—could raise some free speech concerns if done over-aggressively. It could also lead to confusion as definitions shift from administration to administration, with different parties targeting different types of speech. Finally, these efforts could distract from a much bigger problem: new state laws that attempt to impose new DEI-inspired regulations on algorithmic systems.
Definitional Complexities Abound
To start, the terms “misinformation” and “climate change” do not appear in the AI RMF, so it is unclear what this directive from the AI Action Plan is meant to target. As for DEI, these are terms NIST uses to evaluate the governance of LLMs by determining whether they include certain training data and outputs that acknowledge a wide range of views and beliefs; however, these terms are not clearly defined in the AI RMF either. NIST should seek clarity on and definitions for these terms before determining whether they merit inclusion in the auditing process. Importantly, these are just directives for NIST’s evaluation program and do not directly regulate how LLMs are actually built.
A more challenging aspect of the new EO is the “free from top-down ideological bias” requirement for LLMs contracting with the federal government. Again, the terms “top-down” and “bias” lack formal definitions, leading to confusion and leaving this and future administrations wide leeway to interpret the order how they see fit. “[H]aving the government decide what those principles mean in practice is where things always get hairy,” argues Greg Lukianoff, President and CEO of the Foundation for Individual Rights and Expression.
Because neither the Action Plan nor the new EO directly regulate LLM production or use, they are unlikely to alter the trajectory of the LLM market or conflict with the First Amendment. However, they do run counter to the administration’s ethos that the government should not be involved in the speech decisions of private platforms. Allowing the federal government to determine what constitutes “ideological bias,” even for the purposes of contracting, could set a dangerous precedent. While DEI does not necessarily factor into AI training, the government’s attempt to police this appears to be forcing an ideological shift onto the industry. Other parts of the order claim the administration wants AI completely free of ideological bias.
There are practical challenges and costs associated with policing speech in this way. The EO acknowledges that Office of Management and Budget (OMB) guidance on its “Unbiased AI Principles” should “account for technical limitations in complying with this order.” The guidance should also “avoid over-prescription and afford latitude for vendors to comply with the Unbiased AI Principles and take different approaches to innovation.” Such flexibility could help counter the danger of indirect regulatory overreach by some officials or agencies; however, this quasi-regulatory process could still create costs for AI developers, who will be forced to address a new minefield of speech-related concerns.
The More Serious “Woke AI” Problem
A different concern about the new EO is that it could distract from the need to focus on a more serious “woke AI” threat. Some state lawmakers have been considering AI regulatory measures that would formally enshrine DEI-inspired ideas into law. Unlike the new EO, these measures would not be limited to government contracting but would directly regulate private AI development and use.
In May 2024, Colorado Gov. Jared Polis signed into law SB24-205, “Consumer Protections for Artificial Intelligence.” The measure, which will take effect in early 2026, aims to proactively identify and eliminate the potential for “algorithmic discrimination” in some AI systems through preemptive rules. The Multistate AI Policymaker Working Group (MAP-WG), a coalition of state lawmakers from more than 45 states, has attempted to implement similar AI discrimination bills.
These bills take their cues from the Biden administration’s “Blueprint for an AI Bill of Rights,” which was rooted in the belief that algorithmic systems were inherently “unsafe, ineffective, or biased,” and needed to be “designed in an equitable way” through “proactive and continuous measures to protect individuals and communities from algorithmic discrimination” including “proactive equity assessments.” After the Trump administration returned to the White House, MAP-WG members took up this DEI-infused agenda, arguing that “the probability of congressional inaction” on AI regulation means they need to fill the gap with a state patchwork of new rules.
The Trump administration’s new EO does not directly address these measures. However, the AI Action Plan instructs the OMB to “work with Federal agencies that have AI-related discretionary funding programs to ensure, consistent with applicable law, that they consider a state’s AI regulatory climate when making funding decisions and limit funding if the state’s AI regulatory regimes may hinder the effectiveness of that funding or award.”
This would both allow the administration to pressure states against adopting such rules and represent a better approach to dealing with concerns about “woke AI.” To reiterate, the recommendations in the AI Action Plan and new EO focus on private AI models that might exhibit “woke” tendencies. Those private policies are obviously not binding in the same way DEI-inspired state policies would be.
The Administration and Congress Should Preempt “Woke AI” Bills
Congress recently considered an AI regulatory moratorium that would have used federal spending leverage to discourage a patchwork of costly and confusing state AI regulations. Part of the One Big Beautiful Bill, this moratorium would have imposed a 10-year pause on state-level AI regulations. While the measure passed in the House, the Senate ultimately removed it.
Congress still has time to act to stop a state patchwork of “woke AI” bills from developing, and the administration appears more than willing to help. During a speech announcing the AI Action Plan, Trump told the audience, “We need one commonsense federal standard that supersedes all states.” A well-structured national approach is needed now to stop the growth of burdensome AI mandates that interfere with national priorities, including the administration’s efforts to stop “woke AI.”