Brad Smith: the United States oversees AI without defined regulations.

      Microsoft president Brad Smith states that the US is currently regulating AI without a clear framework of rules. He cautions that this ambiguity poses a problem for the entire industry. He made these comments to Fortune during the AI for Good Global Summit.

      “What we currently have is regulation without transparent or comprehensive guidelines,” Smith noted. “In the absence of rules, businesses are unable to strategize.”

      His concerns arise after two sudden actions taken by the Trump administration. Last month, the Commerce Department invoked export-control laws to restrict Anthropic’s Fable 5 and Mythos 5 models globally, citing cybersecurity risks. Shortly thereafter, officials urged OpenAI to postpone the public release of its GPT-5.6 family, granting early access solely to government-approved partners.

      Both restrictions have since been relaxed. Fable 5 was reinstated this month, and GPT-5.6 is now being made available to the public. Smith acknowledges that Washington was justified in addressing the Fable issue but believes the regulatory approach it adopted was flawed.

      “The government received information indicating an urgent cybersecurity threat. When it receives such information, I believe it is appropriate to take action,” he said. “However, it only had one regulatory mechanism at its disposal: an export control.” Legal experts argue that these controls were established before the advent of API-served models, raising questions about their viability in a court of law.

      Critics argue that the outcome resembles a licensing system without any legislative backing. A June executive order initiated a voluntary pre-release review but did not establish formal approval guidelines. The government has yet to disclose who qualifies as a “trusted partner” or which models will be subject to scrutiny next.

      “The government lacks the necessary tools,” Smith stated. “Common sense dictates not to be overly aggressive, but there should be enough sensitivity to take appropriate actions.”

      The situation involving Anthropic has also sparked discussions about sovereign AI—the effort by governments to control the models and their underlying infrastructure. In Europe, a French politician compared the shutdown to a blockade of the Strait of Hormuz, while Canadian Prime Minister Mark Carney remarked that it served as a lesson about relying on too few providers.

      Smith believes that people have misinterpreted the export action as an attempt to isolate foreign entities. He clarifies that the aim was to remove the model from the market entirely. “They requested Anthropic to withdraw Fable, but when Anthropic refused, they resorted to export controls to enforce it, both domestically and internationally,” he explained.

      Regardless, he emphasizes that the responsibility now lies with Washington and US companies to demonstrate that their systems are reliable. “Consumers will only purchase what we offer if they are assured of a consistent supply,” he noted. Salesforce CEO Marc Benioff expressed a contrary view at the same event, stating he felt "good" about the block and believed Europe misunderstood a security measure as an act of aggression.

      The significance of this situation highlights the considerable influence the US currently possesses over which AI models can be distributed globally. However, it has hardly formalized any of this authority. For a business that markets AI on an international scale, this lack of visible guidelines is an even greater concern. While one can debate regulations, dealing with invisible policies poses a much greater challenge for planning.