Trends in the Trump Administration’s approach to AI Governance
| AUTHOR | Agastya Anand |
| DATE | September 10, 2026 |
| CATEGORIES | AI |
From the AI Diffusion Rule to targeted controls
Shortly before the inauguration of Donald Trump as President, the Biden administration issued the ‘AI Diffusion Rule’. The rule divided countries into three tiers governing access to advanced computing chips and in some cases certain closed AI model weights. Tier 1 consisted of the US and 18 trusted partners; Tier 2 encompassed most other countries, including India; while Tier 3 covered Macau and US arms-embargoed destinations including China, Russia, Iran and North Korea. Less than five months later, the Trump administration repealedthe ‘rule’, with an official communique claiming it would have “stifled American innovation”, “saddled companies with burdensome new regulatory requirements”, and “undermined U.S. diplomatic relations with dozens of countries by downgrading them to second-tier status”.
The removal of the rule was not marked by similar policies across the board; and the Trump administration’s policies towards chip exports have betrayed a tension between the desire to stifle Chinese AI development and at the same time to dominate the global AI ecosystem. The US’s decision to block sales of H20 AI chips to China in April 2025, for example, can be seen as a measure aimed at slowing Chinese AI development through the denial of access to technology. However, the restrictions also imposed costs on American companies through lost revenue, reduced developer adoption and the weakening of hardware-software ecosystems such as CUDA. The administration nevertheless subsequently reversed course and resumed licensed H20 exports. In January 2026, BIS shifted licence applications for NVIDIA’s H200, AMD’s MI325X and similar chips from a presumption of denial to case-by-case review subject to security conditions.
China: technology denial or ecosystem competition?
China has emerged as one of the main competitors of the United States in technological advancement. DeepSeek, for example, has emerged as a competitor to leading American AI developers such as OpenAI, Google and Anthropic despite the constraints Chinese companies face in accessing frontier chips.
Specific actions such as America’s restrictions on H20 chip exports have been aimed at slowing down Chinese access to frontier compute and therefore, preserving the US’s technological lead. These export controls are based on a view of the chips as a technological chokepoint for AI development, by constraining compute and therefore AI capability. Chinese progress however, suggests that AI advancement cannot be reduced entirely to hardware capability, and that more restrictive hardware controls may face diminishing returns. Additionally, such export controls can also reduce revenues and scale for US firms trading with China, weaken American software and developer ecosystems within the PRC, and accelerate Chinese substitution.
The issue of chip exports was eventually negotiated alongside several others through US-China trade negotiations, at which the US agreed to continue exports of chips to China. At these trade negotiations, China also allowed for the export of rare earths to the United States. However, the US has still sought to counter Chinese AI advancement through other means.
Pax Silica, a US-led initiative composed of the US and allied nations, is one of these; aiming to “secure” the supply chain, Pax Silica’s focus is on building a trusted supply chain covering critical minerals, semiconductors, advanced manufacturing, compute infrastructure, AI, networks and related technologies. India officially joined Pax Silica this February.
A bloc like Pax Silica presents a trade-off between resilience and fragmentation. On one hand, it will help ensure a secure supply chain for the US made up largely of trusted partners; on the other hand, it can raise costs for its members and may be difficult to sustain in the long run.
Deregulation with national-security exceptions
The inauguration of President Trump was widely welcomed in the tech world, as symbolised by the attendance of multiple major tech CEOs and donations from figures such as Sam Altman. One of the administration’s first AI-related EOs, ‘Removing Barriers to American Leadership in Artificial Intelligence’, also signified a positive boost for AI companies due to its removal of Biden’s AI EO, displaying a broader deregulatory approach.
This, however, did not necessarily manifest in consistently deregulatory measures towards domestic tech companies. Most recently, for example, the blocking of Anthropic’s Mythos and Fable 5 models to non-American users between June and July of this year would make news worldwide; this not only prevented access to the AI model for users outside of the United States, but also for many of Anthropic’s own non-American staff. The episode demonstrated that US export control authorities could be applied directly to access to frontier AI models. The restrictions were subsequently lifted at the end of June.
Another area in which Anthropic and the US government have emerged at odds is regarding the company’s unwillingness to lift safeguards which would allow for the Department of Defense to use it for the purposes of targeting weapons autonomously and domestic surveillance. Anthropic refused to do so, leading to its blacklisting as a “supply-chain risk to national security”, which was struck down by a US judge on August 27.
Anthropic’s blacklisting shows that the administration is willing to intervene in the activities of AI companies on national-security grounds despite its overall deregulatory approach. The court found the supply-chain-risk designation unlawful. The confrontation between the company and the administration appears to have eased since then.
Conclusion
Over the past year, three major themes seem to have appeared in the Trump Administration’s approach to AI governance.
- A deregulatory approach to domestic AI companies, except in cases which the government deems threatening to national security.
- A focus on the global diffusion and commercial dominance of the American AI ecosystem, including securing trusted technology supply chains.
- Usage of targeted restrictions against what are deemed “national security risks”, particularly with regards to China.