By 
A new executive order signed by former U.S. President Donald Trump aims to bar “ideologically biased” artificial intelligence models—labeled as “woke AI”—from use in federal government contracts. The directive, which prioritizes what it calls “ideological neutrality” and “truth-seeking,” could have sweeping implications for how American tech companies train and deploy AI models.
The order follows long-standing concerns among U.S. officials over the ideological alignment of AI systems developed abroad, particularly in China. U.S. security experts and researchers have flagged tools developed by firms like Alibaba and DeepSeek for allegedly avoiding politically sensitive queries and reinforcing Chinese Communist Party narratives. These concerns have fueled calls to develop U.S.-backed AI models that promote democratic values and resist foreign influence.
Now, the Trump administration’s directive shifts the domestic conversation. Signed during an AI-focused event hosted by the All-In Podcast and the Hill & Valley Forum, the order prohibits federal agencies from procuring AI systems that incorporate concepts tied to diversity, equity, and inclusion (DEI), critical race theory, transgender rights, or systemic racism—framing these as partisan ideologies that compromise “accuracy and impartiality.”
“We are getting rid of woke,” Trump declared at the event. “From now on, the U.S. government will only use AI that is fair, objective, and free from ideological distortion.”
The order is accompanied by an “AI Action Plan” that reprioritizes federal AI policy around national competitiveness, defense, and deregulation—explicitly shifting away from a previous focus on social risks and ethics.
Regulatory Shift May Pressure Developers
Industry analysts warn the order may exert significant pressure on AI companies, especially those reliant on federal funding. While executive orders don’t carry the permanence of legislation, federal contracts often serve as critical revenue streams for AI startups navigating high operational costs and investor expectations.
“This order could create a chilling effect,” said Rumman Chowdhury, CEO of tech nonprofit Humane Intelligence and former U.S. science envoy for AI. “Developers may start reshaping training data or filtering outputs not for accuracy or utility, but to align with political mandates.”
The language in the order calls for AI systems that “prioritize historical accuracy, scientific inquiry, and objectivity.” Yet it defines neutrality narrowly, excluding DEI and other frameworks as ideological constructs. Experts argue that this definition is both selective and politically loaded.
Dr. Philip Seargeant, a senior lecturer in linguistics at The Open University, noted that all language—and by extension, all AI output—is inherently shaped by values.
“There is no such thing as a truly neutral AI,” he said. “The idea of pure objectivity in language or model training is a myth.”
Questions of Bias and Enforcement
The order’s timing is significant. Just last week, OpenAI, Anthropic, Google, and xAI all signed contracts with the Department of Defense worth up to $200 million each for the development of AI capabilities to support national security objectives. It remains unclear how—or whether—these companies will adapt to the new directive, and whether models like xAI’s Grok, which leans into contrarian narratives, will be considered compliant.
Despite claims of being “anti-woke,” Grok has made headlines for generating controversial and sometimes offensive content, including antisemitic remarks and historical inaccuracies. Legal scholars like Stanford’s Mark Lemley question whether the order will be applied consistently.
“The real test is whether the government would ban a tool like Grok, which is deliberately engineered to provide politically charged answers,” Lemley said. “If not, it raises concerns about viewpoint discrimination.”
Still, Grok has been added to the General Services Administration schedule, making it accessible to all federal agencies. Meanwhile, statements from xAI founder Elon Musk suggest ambitions to use future versions of the model to “rewrite the entire corpus of human knowledge”—a move critics say would centralize ideological control under the guise of correction.
The Larger Debate: Can AI Ever Be Neutral?
The executive order has ignited broader questions about what it means for an AI system to be “truth-seeking.” In practice, AI models reflect their training data—whether curated from scholarly texts, social media, or web forums—and the philosophical biases of their developers. Even data intended to be neutral can inadvertently favor certain viewpoints through inclusion or omission.
Incidents like Google’s Gemini chatbot depicting racially diverse Nazis have been cited by conservatives as evidence of overcorrection in AI fairness protocols. Trump’s executive order specifically references such outcomes as examples of how DEI priorities can skew reality.
“If AI says climate change is real, is that left-wing bias?” Seargeant asked. “Some argue that objectivity means presenting both sides, even if one lacks scientific merit.”
Conservative voices like David Sacks, now appointed as Trump’s AI czar, argue that current AI leaders lean too heavily into progressive ideologies. On the All-In Podcast, Sacks has repeatedly warned that unchecked “woke AI” threatens free expression and injects political orthodoxy into neutral technologies.
But critics argue the opposite: that such framing enforces a different kind of orthodoxy, one that marginalizes ongoing efforts to build AI that is inclusive, equitable, and socially responsible.
What Comes Next?
While Trump’s order may energize factions pushing for deregulated, contrarian AI development, it also places U.S. companies in a complex legal and ethical position. Developers must now navigate a shifting regulatory landscape while balancing public scrutiny, global competition, and internal commitments to responsible innovation.
In an industry where no model is entirely neutral—and no truth is free from context—the challenge ahead may not be eliminating bias, but managing it transparently.
Comments