US seeks ‘unquestioned’ AI dominance

Red tape

To support the first of these pillars, the US will seek to remove red tape that the government believes is holding back the private sector.

Trump has already made some moves in this regard, including rescinding a previous executive order (EO) issued by former president Joe Biden in 2023 – which attempted to set standards for AI safety and security, protect privacy and spur responsible development. The new administration believes Biden’s EO “foreshadowed an onerous regulatory regime”, and many of the actions laid out in the plan this week reflect a general trend of unpicking his legacy.

It also sets out actions to encourage the use of open source and open weight AI models, enable adoption in critical sectors such as healthcare, empower initiatives to expand AI-enabled education and prepare the workforce for its impact, among other things.

On the second pillar, the plan proposes a number of actions including streamlining the planning and building processes governing datacentre and chip fabrication facilities, as well as expanding and protecting the vast energy infrastructure that will be needed to run AI models.

On the third pillar, it sets out plans to export the full AI technology stack to other countries, lest they turn to rival states such as China. On China specifically, the plan lays down initiatives to counter China’s influence on international governance bodies.

Also contained in the AI action plan are provisions for more AI-focused cyber initiatives, including the potential creation of an AI-specific information and analysis centre (ISAC) at the Department of Homeland Security (DHS), and plans to do more to counter AI model vulnerabilities and fight back against malicious use of AI by threat actors.

No ‘woke’ AI in DC

Alongside the AI action plan and in keeping with its disavowment of diversity, equity and inclusion (DEI) initiatives, the Trump administration also launched a broadside against so-called “woke” AI.

The text of an additional EO signed by Trump on 23 July says large language models (LLMs) should be neutral and nonpartisan tools “that do not manipulate responses in favour of ideological dogmas such as DEI”.

The EO says that in an AI context, this could mean targeting topics such as representation of ethnic or sexual minorities, critical race theory, discrimination on the basis of race or sex, intersectionality, transgender rights and unconscious bias.

The EO’s guidance applies to models used in the federal government, but will likely have ramifications beyond it.

Broadly supportive

Matt Mittelsteadt, research fellow at Washington DC-based libertarian think tank the Cato Institute, was broadly supportive of the action plan’s aims.

“On the positive end, the order includes a much-needed emphasis on the importance of innovation, notably the forthcoming effort to analyse federal rules to identify onerous regulations,” he said. “Also notable is the renewed emphasis on open source and related efforts to enable open source developers.

“The stated emphasis on AI diplomacy could be essential as America’s AI success depends on foreign market access,” said Mittelsteadt. “Finally, the elevated emphasis on cyber security – and specifically enabling cyber defence – is matched to the current state of risk.”

However, he pointed out some sticking points, notably the use of government procurement power to shape model output along ideological lines.

“Not only is ‘objectivity’ elusive philosophically, but efforts to technically contain perceived bias have yet to work,” said Mittelsteadt.

“This is also a mistake from a competitiveness standpoint. If this policy successfully shapes American models, we will lose international customers who won’t want models shaped by a foreign government’s whims.”