The White House has made a bold announcement that marks a new era for federal technology policy by calling for the acceleration of AI development and deployment in America’s national security policies. The order, issued this Friday, could be interpreted as authorizing “responsible” acceleration of AI in the domains of intelligence analysis and military operations, while setting a clear boundary on its usage for “unlawful surveillance and censorship. This is news that comes as a relief to anyone who’s been watching the slow simmer of conflict between Washington, D.C., and Silicon Valley over groundbreaking innovations. That’s a sentiment being felt increasingly inside the policy worlds that the U.S. just can’t wait around in this case any longer, especially with countries it faces challenging times with if it considers using AI as a centerpiece of their strategic planning.
It is very interesting how this directive attempts to balance. On the other hand, the administration’s concern is that its actions will foster the development of private AI systems that will further accelerate the “supply shock” that challenges federal agencies to remain on pace with their commercial counterparts. On the one hand, the administration is urging speed and adoption, recognizing how historically slows its own agencies have been to keep up with commercial AI advances. The other side of the coin, however, is that the memorandum presents a strong case for maintaining constitutional safeguards. The balance between urgency and accountability has not changed, but it’s more acute now, as the pace of AI system change runs faster than existing oversight systems.
The administration had already made a U-turn earlier this week when it invited some of the industry’s most powerful AI models to be among the first to be subject to government-led cybersecurity stress testing before they would be released to the public. This pre-release check is meant to detect flaws that adversary agents can exploit, especially in models that can be used to produce code, to create an attack vector, or to analyse surveillance data. This offer was made on a voluntary basis has received mixed responses from industry watchers. Some view it as a good step for mopping up uncertainty; others believe it might be an indirect way to set up a regulatory system without congressional hearing.
The new stance was codified by President Donald Trump’s national security memorandum. The memorandum states that “Under my Administration, the United States can and will responsibly accelerate the use of AI in the intelligence and war fighting fields in accordance with American values.” The phrasing is important, because it makes it clear that it’s a matter of values and not of ability. The White House recognizes the dangers of putting autonomous or semi-autonomous systems on the battlefield, both in terms of their reputation and as a legal matter, it says.
One of the more tangible parts of the move is Secretary of Defense Pete Hegseth.Secretary of Defense Pete Hegseth is one of the more tangible parts of the move. Trump said Hegseth will have 90 days to consider revoking an existing directive on autonomy of weapons systems. The aim is merely unambiguous: “to ensure that the use of AI systems is deliberate and that they respect the chain of command.” That language recognizes a particular concern of military ethicists and high-ranking officers: Unpredictable AI systems could offer recommendations or actions without human input—without bounds or checks of reason. The Pentagon is getting a message: Don’t let AI take the place of humans in making decisions.
The memorandum, however, carries a significant don’t-do. Trump went on to say that AI technologies should not be created or used in the national security enterprise for the purpose of “censorship of free speech” or conducting “unauthorized or unlawful” surveillance. The provision appears to be directed at domestic, not foreign, military applications of AI. It’s a reminder that although the administration wishes to proceed quickly, it’s not interested in a surveillance state run by black-box algorithms. While civil liberties groups may have reasons for concern, the specific ban on ‘unlawful’ surveillance does give a legal basis to continue to monitor.
The strategic motivation behind this step is more revealing than it might seem.What this move does tell is how strategic leaders have come to realize that AI is not just some hypothetical future experiment. It is already used in satellite image analysis, in threat detection, and in planning logistics, and in Cyber defense. That is the true issue whether federalism can be preserved as it manages AI – without breaking up. From my personal perspective on the topic of scaling new technology in large firms, the biggest challenge is the governance surrounding the new technology, rather than the algorithm itself. The 90-day time period for investing in changing the autonomy of weapons is generous. It may well create a document reminiscent of compliance without real meaningful guardrails, or it may create meaningful guardrails, which remains to be seen.
On the industry side, however, the giants of AI such as Anthropic and other down the frontier model developers are now in a weird situation. Unless cooperation with the voluntary tests is done, it can help to foster trust with Washington. However, potentially overcooperation could present a precedent that inhibits faster release cycles or indirectly provide an indication of proprietary model architecture by testing results. International competitiveness is also a question. As American companies reveal details of pre-release models to government, are they putting themselves at a disadvantage with Chinese competitors or Europe’s, who don’t reveal anything of the kind?
