For 18 months, the Trump administration has been switching back and forth between restricting and allowing different parts of the AI industry — controlling which chips can be sold to China, reviewing powerful AI models for security risks, and banning some of Anthropic's AI products before partially walking those bans back. Politico reported on July 1, 2026, that this approach "could be a gift to China." Politico

The back-and-forth is not accidental. It is built into how the administration is running at least three separate policy efforts at the same time, on different timelines, and sometimes pulling in opposite directions.

Chip Export Controls: The Revolving Door

The clearest example is the rules around selling computer chips to China. These chips, made by a company called Nvidia, are the hardware that powers AI systems. In July 2025, Trump's AI advisor David Sacks defended the decision to loosen restrictions and let Nvidia resume some AI chip sales to China. Bloomberg By August 2025, China was telling its own companies to avoid using Nvidia's H20 chips, especially for government purposes, which made the relaxed rules less useful for Nvidia's business. Bloomberg

By November 2025, the Trump team had discussed selling Nvidia's newer H200 chips to China as a compromise, and by December 2025, Reuters confirmed the US would allow those exports. Bloomberg Reuters At the same time, the Commerce Department was writing rules that would require a permit to ship AI chips anywhere in the world, not just to China. Bloomberg

In January 2026, a separate presidential action adjusted imports of semiconductors and chip-making equipment into the United States, adding tariffs on top of the export-control system. White House

Model Oversight: From Revocation to Review

On his first day in office in 2025, Trump canceled a 2023 Biden order that aimed to reduce AI risks. Reuters By May 2026, the White House was considering requiring government reviews of AI models before companies release them to the public, according to the New York Times. Reuters

In June 2026, Trump signed an executive order called "Promoting Advanced Artificial Intelligence Innovation and Security" that shortened the voluntary review period for powerful AI models from 90 days to 30 days. The administration will ask leading AI developers to voluntarily submit their most capable models for government cybersecurity testing. WSJ Reuters White House

OpenAI, another major AI company, limited access to new models because of government security concerns and said that White House review of AI releases should not become the long-term default. WSJ The tension is clear: the administration calls voluntary submission a security measure, while at least one major developer is already pushing back.

The Anthropic Ban and Its Partial Reversal

The most complicated thread involves Anthropic, an AI company. A ban on Anthropic's Mythos AI model was still in place as of late June 2026 reporting. WSJ By June 26, the administration let Anthropic offer its banned Mythos 5 model again, but only to trusted companies and government partners — a partial rollback, not a full lift. WSJ Cybersecurity executives had urged the administration to ease the restrictions earlier in June. AP News

By July 3, 2026, the administration lifted restrictions on Anthropic's latest versions of its Claude chatbot. AP News

Bloomberg reported on June 26 that the ban on Anthropic's AI may boost China's open-source AI models. Bloomberg "Open-source" means the code is freely available for anyone to use and modify. The logic is simple: if American AI products are restricted, foreign alternatives that are free to use fill the gap.

The China Vector and 'Woke AI'

Running alongside the Anthropic ban is a separate effort targeting tech companies whose AI models are deemed too "woke," as reported in July 2025. WSJ The administration has also vowed to crack down on Chinese companies "exploiting" AI models developed in the United States by using a technique called model distillation — training smaller AI models on the outputs of larger ones, essentially copying what a powerful model has learned. AP News

The White House is considering banning DeepSeek, a Chinese AI company's chatbot, from US government devices due to national security concerns. WSJ

A Trump official said in June 2025 that China is only 3–6 months behind the US in AI capability. Reuters Reuters reported that same month that Trump's approach to AI could be centered on expanding markets abroad for US AI chips and models. Reuters

What This Means for Markets and the Competitive Landscape

The policy pattern matters for anyone with money in AI-focused funds, semiconductor stocks, or big-tech companies. Each reversal or partial rollback changes how much money Nvidia can make from China, how quickly Anthropic can roll out products, and how Chinese open-source models stack up against restricted American ones.

The 30-day review period shortens the time it takes for new AI models to reach the market, but it also sets up a checkpoint that could become a permanent layer of regulation — which is exactly what OpenAI has warned about. The crackdown on model distillation, if enforced, would affect how Chinese firms like DeepSeek learn from US models. But catching this in practice is hard. Distillation is difficult to detect and even harder to prove across different countries' legal systems.

The tension between the "woke AI" effort and the national-security review effort is also worth noting. One set of policies pushes for certain content standards in what AI models produce; another pushes for security checks before release. Developers now face competing demands from the same administration, with no clear rule for which takes priority. The June 2026 executive order tries to combine both into one framework, but the details remain voluntary and vague.

For savers and investors with exposure to AI-themed funds, chipmakers, or big-tech indices, the key thing to watch is not any single order or ban. It is the direction of travel: whether the administration settles on stable rules or keeps switching back and forth. Each switch creates a pricing gap — a sudden change between what markets expected and what policy actually delivers. The 3-to-6-month gap between the US and China, cited by the Trump official, means that every month of uncertainty in US policy gives Chinese competitors another month to catch up by deploying their free alternatives.