Trump’s ‘anti-woke AI’ order could reshape how US tech companies train their models

When DeepSeek, Alibaba, and other Chinese firms released their AI models, Western researchers quickly noticed they sidestepped questions critical of the Chinese Communist Party. U.S. officials later confirmed that these tools are engineered to reflect Beijing’s talking points, raising concerns about censorship and bias.

American AI leaders like OpenAI have pointed to this as justification for advancing their tech quickly, without too much regulation or oversight. As OpenAI’s chief global affairs officer Chris Lehane wrote in a LinkedIn post last month, there is a contest between “US-led democratic AI and Communist-led China’s autocratic AI.”

An executive order signed Wednesday by President Donald Trump that bans “woke AI” and AI models that aren’t “ideologically neutral” from government contracts could disrupt that balance. 

The order calls out diversity, equity, and inclusion (DEI), calling it a “pervasive and destructive” ideology that can “distort the quality and accuracy of the output.” Specifically, the order refers to information about race or sex, manipulation of racial or sexual representation, critical race theory, transgenderism, unconscious bias, intersectionality, and systemic racism. 

Experts warn it could create a chilling effect on developers who may feel pressure to align model outputs and datasets with White House rhetoric to secure federal dollars for their cash-burning businesses. 

The order comes the same day the White House published Trump’s “AI Action Plan,” which shifts national priorities away from societal risk and focuses instead on building out AI infrastructure, cutting red tape for tech companies, shoring up national security, and competing with China. 

The order instructs the director of the Office of Management and Budget along with the administrator for Federal Procurement Policy, the administrator of General Services, and the director of the Office of Science and Technology Policy to issue guidance to other agencies on how to comply. 

“Once and for all, we are getting rid of woke,” Trump said Wednesday during an AI event hosted by the All-In Podcast and Hill & Valley Forum. “I will be signing an order banning the federal government from procuring AI technology that has been infused with partisan bias or ideological agendas, such as critical race theory, which is ridiculous. And from now on the U.S. government will deal only with AI that pursues truth, fairness, and strict impartiality.”

Determining what is impartial or objective is one of many challenges to the order.

Philip Seargeant, senior lecturer in applied linguistics at the Open University, told TechCrunch that nothing can ever be objective. 

“One of the fundamental tenets of sociolinguistics is that language is never neutral,” Seargeant said. “So the idea that you can ever get pure objectivity is a fantasy.”

On top of that, the Trump administration’s ideology doesn’t reflect the beliefs and values of all Americans. Trump has repeatedly sought to eliminate funding for climate initiatives, education, public broadcasting, research, social service grants, community and agricultural support programs, and gender-affirming care, often framing these initiatives as examples of “woke” or politically biased government spending. 

As Rumman Chowdhury, a data scientist, CEO of the tech nonprofit Humane Intelligence, and former U.S. science envoy for AI, put it, “Anything [the Trump administration doesn’t] like is immediately tossed into this pejorative pile of woke.”

The definitions of “truth-seeking” and “ideological neutrality” in the order published Wednesday are vague in some ways and specific in others. While “truth-seeking” is defined as LLMs that “prioritize historical accuracy, scientific inquiry, and objectivity,” “ideological neutrality” is defined as LLMs that are “neutral, nonpartisan tools that do not manipulate responses in favor of ideological dogmas such as DEI.”

Those definitions leave room for broad interpretation, as well as potential pressure. AI companies have pushed for fewer constraints on how they operate. And while an executive order doesn’t carry the force of legislation, frontier AI firms could still find themselves subject to the shifting priorities of the administration’s political agenda.

Last week, OpenAI, Anthropic, Google, and xAI signed contracts with the Department of Defense to receive up to $200 million each to develop agentic AI workflows that address critical national security challenges. 

It’s unclear which of these companies is best positioned to gain from the woke AI ban, or if they will comply. 

TechCrunch has reached out to each of them and will update this article if we hear back. 

Despite displaying biases of its own, xAI may be the most aligned with the order — at least at this early stage. Elon Musk has positioned Grok, xAI’s chatbot, as the ultimate anti-woke, “less biased,” truthseeker. Grok’s system prompts have directed it to avoid deferring to mainstream authorities and media, to seek contrarian information even if it’s politically incorrect, and to even reference Musk’s own views on controversial topics. In recent months, Grok has even spouted antisemitic comments and praised Hitler on X, among other hateful, racist, and misogynistic posts. 

Mark Lemley, a law professor at Stanford University, told TechCrunch the executive order is “clearly intended as viewpoint discrimination, since [the government] just signed a contract with Grok, aka ‘MechaHitler.’” 

Alongside xAI’s DOD funding, the company announced that “Grok for Government” had been added to the General Services Administration schedule, meaning that xAI products are now available for purchase across every government office and agency. 

“The right question is this: would they ban Grok, the AI they just signed a large contract with, because it has been deliberately engineered to give politically charged answers?” Lemley said in an email interview. “If not, it is clearly designed to discriminate against a particular viewpoint.”

As Grok’s own system prompts have shown, model outputs can be a reflection of both the people building the technology and the data the AI is trained on. In some cases, an overabundance of caution among developers and AI trained on internet content that promotes values like inclusivity have led to distorted model outputs. Google, for example, last year came under fire after its Gemini chatbot showed a black George Washington and racially diverse Nazis — which Trump’s order calls out as an example of DEI-infected AI models.  

Chowdhury says her biggest fear with this executive order is that AI companies will actively rework training data to tow the party line. She pointed to statements from Musk a few weeks prior to launching Grok 4, saying that xAI would use the new model and its advanced reasoning capabilities to “rewrite the entire corpus of human knowledge, adding missing information and deleting errors. Then retrain on that.”

This would ostensibly put Musk into the position of judging what is true, which could have huge downstream implications for how information is accessed. 

Of course, companies have been making judgment calls about what information is seen and not seen since the dawn of the internet. 

Conservative David Sacks — the entrepreneur and investor who Trump appointed as AI czar — has been outspoken about his concerns around “woke AI” on the All-In Podcast, which co-hosted Trump’s day of AI announcements. Sacks has accused the creators of prominent AI products of infusing them with left-wing values, framing his arguments as a defense of free speech, and a warning against a trend toward centralized ideological control in digital platforms.

The problem, experts say, is that there is no one truth. Achieving unbiased or neutral results is impossible, especially in today’s world where even facts are politicized. 

“If the results that an AI produces say that climate science is correct, is that left wing bias?” Seargeant said. “Some people say you need to give both sides of the argument to be objective, even if one side of the argument has no status to it.”