Trump’s ‘inclusive AI’ orders draw expert criticism

US President Donald Trump has signed three executive orders on artificial intelligence. They set out practices and strategies intended to create a more inclusive environment for the technology’s development.

The first, PREVENTING WOKE AI IN THE FEDERAL GOVERNMENT, asserts that the federal government must forgo AI models that “sacrifice truthfulness and accuracy for ideological ends”.

“LLMs shall be neutral, nonpartisan tools that do not skew answers in favour of ideological dogmas such as DEI,” the text says.

Developers are prohibited from deliberately integrating ideological judgments into AI responses unless a user’s prompt explicitly calls for it.

“Once and for all, we are getting rid of ‘woke’. I will sign an order banning the federal government from procuring AI technologies that have partisan or ideological tenets built in, such as critical race theory — it is absurd. From this moment on, the US government will use only AI that strives for truth, fairness and strict impartiality,” Trump said.

AI is increasingly embedded in Americans’ everyday lives and plays an important role in how people gain knowledge and information. That makes the reliability of its conclusions especially consequential.

Other initiatives include:

• an order aimed at spurring innovation and development by scrapping so‑called “burdensome federal rules that hinder the development and deployment of AI”;
• a third document — aimed at creating and executing the “American AI Export Programme” to support the development and deployment of an American AI technology stack abroad.

What counts as impartial and objective?

Philip Seargeant, a senior lecturer in applied linguistics at the Open University, noted that nothing can be truly objective.

“One of the fundamental principles of sociolinguistics is that language is never neutral. So the idea of achieving pure objectivity is a fantasy,” he said.

The Trump administration’s ideology does not always reflect the beliefs and values of all Americans. The president has repeatedly sought to cut funding for climate initiatives, education, public broadcasting, scientific research, social programmes, support for rural communities and gender‑affirming care, often presenting these as examples of “woke”.

“Everything that [the Trump administration] does not like gets immediately consigned to the pejorative category ‘woke’,” said data specialist Rumman Chowdhury, head of the non‑profit Humane Intelligence and a former US special envoy on AI.

“Truth‑seeking” is taken to mean that LLMs “prioritise historical accuracy, scientific inquiry and objectivity”, and “ideological neutrality” that models are “neutral, nonpartisan tools that do not skew answers in favour of ideological dogmas such as DEI”.

Elon Musk casts Grok as an alternative to the “woke” approach — less biased and oriented toward the pursuit of truth. The model’s system prompts recommend avoiding references to mainstream media and official sources, seeking opposing viewpoints — including politically incorrect ones — and taking the entrepreneur’s own position into account on contentious issues.

Recently, Grok has been embroiled in scandals over antisemitic and other provocative statements.

“Obviously, the order is aimed at viewpoint discrimination, given that [the government] just signed a contract with Grok, also known as ‘MechaHitler’,” noted Stanford law professor Mark Lemley.

Together with funding from the Department of Defense, xAI also said that “Grok for Government” had been added to the General Services Administration schedule. That means xAI’s products are now available for purchase by all federal departments and agencies.

“The right question is this: will they ban Grok — the AI with which they just signed a big contract — for being deliberately designed to deliver politically coloured answers? If not, this is clearly an attempt to suppress a particular point of view,” Lemley said.

Chowdhury’s chief concern is that, because of the order, companies may deliberately alter training datasets to align with administration policy. She recalled statements by Musk a few weeks before the launch of Grok 4, in which he claimed that xAI “will rewrite the entire corpus of human knowledge, adding missing information and eliminating errors”.

Experts said there is no single objective truth. Achieving fully impartial or neutral outcomes is impossible, especially in today’s world where even facts become the subject of political dispute.

“If an AI produces the result that climate science is correct — does that count as left‑wing bias? Some say you have to represent both sides of an argument to be objective, even when one of the sides has no scientific value,” Seargeant asked.

A plan to win the AI race

The measures are part of the administration’s initiative “Winning the AI Race: America’s AI Action Plan” (Winning the AI Race: America’s AI Action Plan). It sets out 90 federal policies across three tracks: accelerating innovation, building infrastructure, and leading in international diplomacy and security.

“America is currently the world leader in building data centres, hardware performance and the creation of AI models. It is essential that the United States use this advantage to forge a resilient global alliance, preventing our adversaries from taking advantage of our innovations and investments,” the document says.

It mentions the need to tighten controls on AI‑chip exports through “creative approaches” and offers two policy proposals:

• a call for government bodies to work with the AI industry on chip geolocation‑verification features;
• a recommendation to create an enforcement mechanism for future export restrictions.

The plan was drafted by the administration’s technology and AI team, which includes representatives of Silicon Valley, among them Office of Science and Technology director Michael Kratsios, AI and cryptocurrency lead David Sacks and National Security Adviser Marco Rubio. More than 10,000 stakeholders submitted comments that were taken into account in compiling the document.

The plan’s main pillars:

• large‑scale AI‑infrastructure build‑out: constructing data centres, chip‑fabrication plants and power sources, using federal land, bypassing environmental constraints;
• deregulation: curbing state‑level AI rules, tying federal funding to a rejection of regulations, surveying business and the public about obstructive rules;
• protecting free speech and combating ‘biased AI’: excluding DEI, disinformation and climate from government risk assessments, requiring objectivity from models in public procurement;
• support for open models: expanding compute access for start‑ups and researchers, partnering with major developers, supporting projects such as Meta, Hugging Face and AI2;
• AI safety: research into interpretability, control systems and robustness to attacks, hackathons at the Department of Defense and Department of Energy to identify vulnerabilities, assessing risks of using the technology in cyberattacks and weapon creation, cooperation between developers and government bodies to assess threats;
• curbs on China: identifying AI models that reflect party censorship, gathering intelligence on developments in “hostile” countries, countering the inflow of chips and other products from the PRC into the United States;
• national security: integrating artificial intelligence into US defence and intelligence structures, building data centres for the Pentagon, workforce training, process automation, priority access to compute in a crisis.

In July, the Trump administration revised its position on the import of AI chips into China. It allowed Nvidia to sell processors because the company “is not transferring the best technologies”.