Summary of Key Points
The U.S. approach to AI regulation has shifted from a complete laissez-faire attitude to a more cautious and limited approach: The Trump administration previously reversed previous regulatory measures and removed obstacles to the development of AI. Now, the government requires cutting-edge AI models to undergo a 30-day voluntary review before release. Specific restrictions have been imposed on companies like Anthropic (which had two of its models taken offline) and OpenAI (whose GPT-5.6 launch was limited in scope). Additionally, internal divisions within the White House—between the security faction, the innovation faction, and the compromise faction—have influenced policy decisions. The government is also considering investing in leading AI companies, with proposals ranging from a 5% stake to a 50% ownership stake, in an attempt to balance innovation, security, and public interests.
Detailed Explanation
1. A U-Turn in U.S. AI Regulation
During Trump's tenure, the approach to AI was relatively laissez-faire: Upon taking office in 2025, he revoked previous executive orders requiring AI companies to conduct safety tests and share results, relaxed data center regulations, and even asked the Department of Commerce to review state-level AI regulations deemed overly burdensome, aiming to remove any obstacles hindering the advancement of AI in the U.S.
However, the tide has turned this year. A proposed 90-day review period for models was temporarily canceled due to opposition from Silicon Valley leaders like Elon Musk; it was later revised to a 30-day voluntary review period (allowing developers to prepare simultaneously without delaying releases), with the clear statement that mandatory licensing would not be implemented. Trump himself stated, "We need safeguards, but we should interfere as little as possible and act quickly when danger arises." This represents a shift from complete non-intervention to setting up a safety net without overly restricting innovation.
2. Internal Divisions within the White House
The policy change reflects a struggle among three factions:
- Security Faction (led by Secretary of Defense Hagels): Concerned that advanced AI could be weaponized by foreign adversaries (e.g., for cyberattacks or biological warfare), they advocate for strict regulation. In February, they classified Anthropic as a "supply chain risk" and urged the government to disable its technology, although this order was later overturned by court.
- Innovation Faction (led by tech advisor Sachs): Supports a competitive environment and believes that excessive regulation will hinder U.S. AI development. The cancellation of the 90-day review period in May was the result of their lobbying efforts, and they viewed the revised 30-day review as a change in the rules of the game.
- Compromise Faction (led by Chief of Staff Wilson and Secretary of the Treasury Bennett): Seeking a middle ground, proposing that companies voluntarily provide the government with access to new models to ensure safety without stifling innovation. The final policy issued in June reflects this compromise.
3. Regulation in Action
These policies are not just theoretical; there have been concrete instances:
- Anthropic: In mid-June, the Department of Commerce imposed export restrictions on its Fable5 model due to security concerns (the possibility of malicious users unlocking sensitive functions). Anthropic eventually agreed to additional safeguards: the system will automatically block attempts to unlock harmful features and switch to an older version.
- OpenAI: At the end of June, OpenAI was required to limit the initial release of GPT-5.6 to trusted partners, with each access request requiring approval from government officials. This represents direct government intervention in the rollout of cutting-edge models.
4. Government Interest in AI Companies
In addition to regulation, the U.S. government is also interested in gaining a financial stake in AI companies:
- OpenAI’s 5% Stake Proposal: CEO Altman discussed the idea of companies contributing 5% of their shares to a fund similar to Alaska’s Permanent Fund (which uses oil revenues to benefit residents). Based on OpenAI’s valuation of $85.2 billion, this would amount to $42.6 billion; however, this is still in the conceptual stage and requires congressional legislation.
- Trump’s Previous Efforts: He suggested the government could invest 9.9% in Intel in 2025, with the stock price subsequently rising significantly. Trump even expressed the belief that a larger stake would have been more beneficial.
- Sanders’ Radical Proposal: Senator Sanders proposed that the government should hold a 50% stake in major AI companies to distribute profits to all Americans. OpenAI itself has discussed the idea of a "public wealth fund" to share AI benefits with the public.
5. Challenges in Global Regulation
Basso from the World Economic Forum noted that countries are struggling to find the right balance in AI regulation, with political considerations often overshadowing scientific arguments. She pointed out that good regulation can foster innovation by setting safety standards that build user trust and encourage investment. The challenge lies in determining what constitutes effective regulation—too strict measures can stifle innovation, while too lenient policies can create risks. She also emphasized the importance of international cooperation to achieve competitive outcomes.
Conclusion
U.S. AI regulation is currently a delicate balancing act: it must protect against security threats and technological vulnerabilities while fostering innovation to prevent being outpaced by competitors like China. The future direction will depend on the outcome of internal debates, Silicon Valley’s response, and congressional support for investment initiatives. Other countries around the world are closely watching U.S. practices, as AI is a key driver of future competition.