President Trump is unlikely to support the creation of a dedicated federal AI regulator in the United States, according to Sriram Krishnan, an outgoing tech adviser who recently spoke with the Financial Times. Krishnan’s assessment points to a familiar fault line in American technology policy: whether AI should be treated primarily as an innovation engine best guided by industry norms, or as a fast-moving general-purpose capability that requires enforceable government oversight.
What makes Krishnan’s comments notable is not only the conclusion—“never” backing a US AI regulator—but the reasoning behind it. In his telling, Trump’s position is rooted less in a momentary political calculation and more in a broader skepticism toward government intervention. That skepticism, Krishnan suggests, is likely to persist even as public concern about AI risks grows louder and more organized. The implication is that the US may continue to pursue a regulatory posture that looks more like voluntary compliance, sector-by-sector guardrails, and executive-branch guidance than a single, centralized agency empowered to set binding rules for AI systems.
To understand why this matters, it helps to look at what “AI regulation” has come to mean in the American debate. For some advocates, regulation is a way to prevent foreseeable harms: discriminatory outcomes, unsafe deployment in high-stakes settings, privacy violations, and the misuse of AI for fraud or manipulation. For others, regulation is seen as a brake on innovation—especially when the technology evolves faster than bureaucratic processes can respond. In the US, where the political culture often treats regulation as a tradeoff against economic dynamism, the question is rarely just technical. It is also ideological: how much power should the state have over a domain that companies believe they can manage through internal controls?
Krishnan’s remarks land squarely in that ideological space. He frames Trump’s stance as consistent rather than reactive. That distinction is important. If the president’s view were merely a response to current headlines, it might shift if the political environment changes. But if the view is anchored in a deeper preference for limited government involvement, then even escalating backlash may not produce the kind of policy pivot that many observers are hoping for.
The timing of the comments also signals something about the direction of travel in US AI governance. Public pressure around AI has intensified across multiple fronts: concerns about job displacement, fears of deepfakes and election interference, worries about surveillance and data exploitation, and anxiety about safety in systems that can behave unpredictably. These concerns have moved from niche policy circles into mainstream politics and culture. Yet Krishnan’s message suggests that the White House may interpret this pressure differently than critics do. Instead of concluding that a regulator is necessary, the administration may conclude that the best path is to keep policy flexible, avoid creating a new bureaucracy, and rely on existing legal frameworks—consumer protection, civil rights enforcement, product safety rules, and sector-specific regulations—to address harms as they arise.
That approach has supporters. They argue that AI is not a single product category; it is a capability embedded in countless products and services. A one-size-fits-all regulator could either be too blunt to be effective or too slow to keep up with rapid iteration. They also point out that the US already has a patchwork of laws that can apply to AI-related harms, even if those laws were not written specifically for machine learning. In their view, the real challenge is not the absence of regulation but the clarity of enforcement and the speed at which regulators can adapt.
But critics counter that the patchwork model often fails in practice. When harms are novel—such as automated decision systems that are difficult to audit, or generative models that can produce persuasive misinformation at scale—existing laws may be hard to apply consistently. Enforcement can become reactive, case-by-case, and uneven across jurisdictions. A dedicated AI regulator, they argue, could establish baseline requirements for transparency, risk management, testing, and accountability—requirements that would apply before harm occurs rather than after it becomes visible.
Krishnan’s comments suggest that the White House is not persuaded by that argument. If Trump is unlikely to back a regulator, then the burden of governance shifts elsewhere: to Congress, to state governments, to agencies with existing mandates, and to private-sector standards. That shift could reshape the landscape in ways that are less visible than a new federal agency but still consequential.
One likely outcome is continued emphasis on “guardrails” without a single central authority. In the US, this often takes the form of executive-branch initiatives, voluntary frameworks, and guidance documents that encourage companies to adopt best practices. These efforts can be useful, especially when they focus on concrete operational steps—like documenting training data sources, conducting red-team testing, implementing monitoring for misuse, and establishing incident response procedures. However, voluntary frameworks can also create a two-tier system: companies that invest early in compliance may gain reputational advantages, while others may delay until enforcement becomes unavoidable.
Another possibility is that the US leans more heavily on sector-specific oversight. AI is already present in healthcare, finance, education, hiring, transportation, and advertising. Each of these sectors has its own regulatory ecosystem. If the federal government resists a unified AI regulator, it may still increase scrutiny through existing channels—requiring more documentation, demanding stronger risk assessments, or imposing penalties when AI systems violate consumer protection or anti-discrimination laws. This could lead to a fragmented regulatory experience for companies, but it might also allow regulators to tailor requirements to the specific risks of each domain.
There is also the question of international alignment. Europe’s approach to AI regulation has been widely discussed, particularly because it includes enforceable obligations tied to risk categories. If the US does not create a comparable framework, American companies operating globally may still need to comply with European rules to access that market. That reality could indirectly shape US behavior: even if Washington resists a regulator, companies may adopt stricter internal controls to satisfy foreign requirements. Over time, this can blur the practical difference between “regulated” and “not regulated,” because compliance becomes driven by market access rather than domestic law.
Yet Krishnan’s comments imply that the political leadership may prefer to avoid the symbolic and institutional step of creating a regulator. Symbolism matters in US policy. A dedicated AI regulator would represent a major expansion of federal authority into a domain that many voters associate with innovation and national competitiveness. It could also become a focal point for partisan conflict: supporters might see it as necessary protection, while opponents might portray it as bureaucratic overreach that stifles American leadership in AI.
In that context, Krishnan’s framing of Trump’s stance as anti-intervention reads like a warning to those expecting a dramatic policy reversal. Even if the public backlash grows, the administration may treat the backlash as evidence that communication and voluntary responsibility are needed—not necessarily that the government should take direct control.
Still, “no regulator” does not mean “no policy.” It means the policy will likely be distributed across other mechanisms. One of the most important mechanisms is procurement and contracting. Governments can influence behavior by requiring vendors to meet certain standards. If the federal government buys AI systems—whether for cybersecurity, public services, or internal operations—it can demand documentation, testing results, and safeguards. This can function as a de facto regulatory regime without creating a new agency. Companies that want access to government contracts may adopt compliance practices that exceed what they would otherwise do.
Another mechanism is enforcement of existing laws. Regulators can pursue cases involving deceptive practices, privacy violations, discrimination, or unsafe products. While this is not the same as proactive risk management, it can still change corporate incentives. If enforcement becomes more frequent and more targeted at AI-specific harms, companies may treat compliance as a necessity rather than a courtesy.
A third mechanism is standard-setting by industry and civil society. Even without a federal regulator, the US can develop norms through technical standards, auditing practices, and certification schemes. These can be influential, especially when they become embedded in procurement requirements or insurance underwriting. The downside is that standards can vary in quality and may be captured by industry interests unless there is meaningful participation from independent experts and affected communities.
Krishnan’s comments also raise a subtler question: what does “AI regulator” mean in the first place? In public discourse, it often refers to a single agency with broad authority. But there are intermediate models—such as a commission, a council with enforcement powers, or a regulator housed within an existing department. The political resistance might be directed not only at the concept of regulation but at the creation of a new institution. If so, the administration could still support some form of oversight structure that does not look like a traditional regulator.
However, Krishnan’s wording suggests a more categorical opposition. If the president “will never” back a regulator, then even intermediate models may face resistance. That would leave the US with a governance approach that is more incremental and less centralized.
For companies building AI systems, the practical impact is straightforward: planning for a future where a single federal regulator sets comprehensive rules may be less realistic. Instead, they may need to prepare for a patchwork of requirements—some coming from enforcement actions, some from sector regulators, some from state-level initiatives, and some from international compliance demands. That can increase compliance costs and uncertainty, but it also encourages companies to build robust internal governance systems that can adapt to different expectations.
For consumers and workers, the implications are more complex. Centralized regulation can provide clearer protections and more predictable enforcement. Without it, protections may depend on the ability of existing institutions to keep pace with AI-driven harms. That can be challenging when AI systems evolve quickly and when harms are difficult to attribute. For example, if an AI system produces biased outcomes, determining responsibility may require technical audits and data access that are not always available. If a generative model is used to create convincing scams, proving causation and identifying the responsible party can be difficult. In such cases, a regulator with specialized expertise could help. Without one, the burden may fall on litigation, complaints, and enforcement agencies that may not have the same technical capacity.
This is where Krishnan’s