Technological companies that seek to sell their artificial intelligence technology to the federal government must now face a new regulatory obstacle: to prove that their chatbots are not “awake”.
President Donald Trump’s new radical plan to counter China to achieve “world domination” in AI promises to reduce regulations and cement American values in AI tools increasingly used at work and at home.
But One of the three executive orders of Trump AI Signed Wednesday – that “preventing an AI of awakening in the federal government” – marks the first time that the American government explicitly tried to shape the ideological behavior of the AI.
Several main suppliers of IA language models targeted by the command – products like Google’s Gemini and the Microsoft co -pilot – have so far been silent on Trump’s anti -revealing directive, which has always faced a study period before embarking on the official supply rules.
While the technology industry has largely welcomed the wider plans of Trump AI, the anti -Swé order forces industry to jump in a cultural war battle – or to do its best to avoid it quietly.
“It will have a massive influence in the industry at the moment”, especially since technological companies already capitulate other directives of the Trump administration, said the defender of civil rights Alejandra Montoya-Boyer, main director of the Center for Civil Rights and Technologies of the Leadership Conference.
This decision also pushes the technological industry to abandon omnipresent forms of racial and gender bias These studies and examples of the real world have proven to be cooked in AI systems.
“First of all, there is nothing like Woke IA,” said Montoya-Boyer. “There is AI technology that discrimination, then there is AI technology that really works for everyone.”
Mold behaviors Models of great language of AI are difficult because of the way they are built and the inherent randomness of what they produce. They were trained On most of what is on the Internet, reflecting the prejudices of all the people who have published comments, published a Wikipedia entrance or images shared online.
“It will be extremely difficult for technological companies to comply,” said former Biden administration official Jim Secreto, who was a deputy chief of staff to the US Secretary of Commerce Gina Raimondo, architect of numerous AI industry initiatives by President Joe Biden. “The models of large languages reflect the data on which they have formed, including all contradictions and prejudices in human language.”
Technology workers also have a say on how they are designed, from the The global workforce of annotators Who check their answers to the Silicon Valley engineers who perform the instructions on how they interact with people.
Trump’s order targets these “descending” efforts in technological companies to incorporate what she calls the “destructive” ideology of diversity, equity and inclusion in AI models, including “concepts such as the critical theory of race, transgender, unconscious prejudice, intersectionality and systemic racism”.
The directive invited to comparison with the heavier efforts of China to ensure that the generative tools of the AI reflect fundamental values of the leading communist party. Secreto said the order looks like a China game book in “the use of state power to eliminate what it considers disadvantaged points of view”.
The method is different, China based on direct audit regulations of AI models, by approving them before their deployment and forcing them to filter the prohibited content as the Bloody Tiananmen Square Crackdown On pro-democracy manifestations in 1989.
Trump’s order does not call such filters, based on technological companies to show instead that their technology is ideologically neutral by disclosing some of the internal policies that guide chatbots.
“The Trump administration takes a softer but coercive route using federal contracts as a lever,” said Secreto. “This creates strong pressure for companies to be self -censors in order to stay in the good graces of the government and to cause the money.”
The call of order to the “search for truth” echoes the language of the president’s unique ally and advisor Elon Musk, who made it the mission of the Grok chatbot carried out by his company XAI.
But that Grok or his rivals will be favored by the new policy remains to be seen.
Despite an introductory “rhetorical” introduction, exposing the problems of the Trump administration with Dei, the real language of the order guidelines should not be difficult for technological companies, said Neil Chilson, a former republican chief technology for the Federal Trade Commission.
“This does not even prohibit an ideological program”, just that all the intentional methods to guide the model be disclosed, said Chilson, responsible for AI policy at the Institute of Non -Lucrative Abundance. “Which is quite light, frankly.”
Chilson challenges comparisons with the censorship china censorships of AI censorship.
“There is nothing in this order that says that companies must produce or cannot produce certain types of production,” he said. “He indicates that promoters do not intentionally access partisan or ideological judgments.”
With their AI tools already widely used in the federal government, technological companies have reacted with caution. Openai said on Thursday that he is expecting more detailed advice, but believes that his work to do the objective of Chatgpt already makes technology consistent with the Trump directive.
Microsoft, a large online service provider to the government, refused to comment.
The XAI of Musk, through the spokesperson Katie Miller, a former Trump official, underlined a comment from the company praising the announcements of Trump AI but did not address the order of purchases. XAI recently announced that he had received an American defense contract up to $ 200 million, just a few days after Grok had publicly published an anti -Semitic comment dam that praised Adolf Hitler.
Anthropic, Google, Meta and Palantant did not respond to requests for comments sent by e-mail on Thursday.
The ideas behind the Order completed for more than a year on the podcasts and the social media flows of the best adviser to Trump’s AI, David Sacks and other influential capital venture capital of Silicon Valley, many of which approved Trump’s presidential campaign last year. Their ire was centered on the Google version in February 2024 of an AI image generation tool which produced historically inaccurate images before the technology giant Removed and repaired the product.
Google later explained that errors – in particular the generation of portraits of black, Asian and Amerindian men when they were asked to show the American founding fathers – was the result of overcompensation for technology which, left to itself, was subject to favoring people with lighter skin due to an omnipresent bias in systems.
Trump’s allies allegedly alleged that Google engineers hardly cod their own social program in the product.
“It is 100% intentional,” said Marc Andreessen, an eminent capitalist venture capitalist and advisor to Trump, on a podcast in December. “This is how you get Black George Washington at Google. There is a replacement in the system that says essentially, literally:” Everyone must be black. “Boom.
Sacks has credited a conservative strategist that fought Dei’s initiatives In colleges and workplaces to help write the order.
“When they asked me how to define” awake “, I said that there was only one person to call: Chris Rofo. And now it is the law: the federal government will not buy Wokeai,” wrote Sacks on X.
Rofo replied that he had helped “identify the ideologies of the DEI in the operational constitutions of these systems”.
But some who agreed that Biden went too far by promoting Dei also fears that Trump’s new order was not a bad precedent for future government efforts to shape AI policy.
“The idea of carrying out an ideological neutrality with AI models is really inrourse,” said Ryan Hauser of the Mercatus Center, a group for reflection on the free market. “And what do we get?” We obtain these border laboratories which change their speeches to meet the political requirements of the moment. ”
