Microsoft’s Brad Smith criticizes US AI regulation for lack of clarity

Microsoft President Brad Smith has a message for Washington: you can’t regulate an industry with vibes. Speaking at the AI for Good Global Summit on July 9, Smith called the current US approach to AI policy “regulation without transparent or complete rules,” a phrase that sounds diplomatic but translates roughly to “nobody knows what’s legal anymore.”

The criticism lands at a particularly awkward moment for the Trump administration, which in June directed the Commerce Department to slap export controls on Anthropic’s Fable 5 and Mythos 5 AI models over cybersecurity concerns. Those restrictions were lifted in early July, but the damage to business confidence was already done.

Export controls as a blunt instrument

Here’s the thing about Smith’s complaint: it’s not that he opposes AI regulation. He’s been one of Big Tech’s more vocal advocates for structured AI governance. Microsoft published a detailed five-point blueprint for public AI policy back in 2023. The issue is that the government appears to be improvising.

The Anthropic situation is a case study in regulatory whiplash. In June, the Commerce Department decided that Fable 5 and Mythos 5 posed enough of a security risk to restrict them worldwide. A few weeks later, the controls were gone.

Smith also referenced restrictive measures affecting OpenAI’s GPT-5.6 rollout, suggesting the problem extends well beyond a single company.

Smith’s core argument is that export controls are fundamentally ill-suited for frontier AI models. The government, he suggests, simply doesn’t have the right tools for the job.

Why this matters beyond Silicon Valley

The regulatory fog around AI isn’t just a headache for Microsoft and its peers. It has real implications for capital allocation across the entire tech sector. When companies can’t predict whether their products will be restricted, approved, or somewhere in between, investment decisions get delayed.

The governance gap

Smith’s 2023 blueprint called for a structured approach to AI policy, one that would give companies clear guardrails rather than surprise enforcement actions.

The fundamental tension is between speed and deliberation. AI capabilities are advancing faster than any regulatory body can reasonably keep up with. But the current approach of reactive, ad hoc enforcement actions isn’t really regulation at all.

There’s a legitimate national security dimension to frontier AI models. The question is whether the tools being used, primarily export controls designed for semiconductors and weapons systems, are the right ones for software that evolves on a weekly basis.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.