Translation result.
The political tide around artificial intelligence is shifting. The Trump administration has moved away from a posture that emphasized resisting and dismantling AI regulation and is now leaning toward a federal licensing regime for AI models. In Congress, support for regulating AI is growing among lawmakers from both parties.
If Democrats reclaim either chamber in November, the passage of some form of federal AI legislation is almost certain. President Trump’s recent trip to China also signaled a notable shift in the administration’s approach to international AI governance.
Two forces are driving this change. First is public backlash. Concern about AI has swelled in recent months, a trend the Wall Street Journal spotlighted this week. A viral clip of Arizona graduates booing former Google CEO Eric Schmidt whenever he mentioned AI at a commencement is one vivid example.
Polls show most Americans feel more fear than optimism about AI, with gaps in some surveys reaching about 40 percentage points. Seven in 10 oppose building data centers near their communities. An Annenberg Public Policy Center survey released last week found two-thirds of Americans believe the government has been too passive on AI regulation — a view shared by Republicans, independents and a majority of Democrats.
No politician can ignore that sentiment. Trump, who is sensitive to public mood, has taken notice. Populist, politically astute aides such as Treasury Secretary Scott Bessent and Susie Wiles have wrested AI policy influence from Silicon Valley “tech-bro” advisers like David Sacks, Sriram Krishnan and Michael Kratsios.
Those steeped in venture capital often miss how ordinary people perceive AI and how badly the White House could damage the Republican Party at the ballot box if it mishandles the issue.
OpenAI’s decision to endorse both the bipartisan Children’s Online Safety Act (KOSA) and Illinois’ SB 315 illustrates this shift. KOSA would require online platforms that children might use to implement measures to prevent and mitigate harm.
SB 315 would require companies developing frontier AI models to build safety systems, undergo annual audits, report major incidents and protect whistleblowers. Many tech trade groups have lobbied against KOSA as vague and potentially unconstitutional. OpenAI previously opposed a California bill similar to SB 315; its turnaround suggests it, too, has recognized the changing political winds.
The second force is Mythos. Anthropic’s powerful model alarmed governments with capabilities that resembled superior hacking performance. It reminded officials that AI is a potent dual-use technology and that private developers cannot unilaterally decide when, how or to whom to release systems. The Trump administration reportedly effectively vetoed Anthropic’s plans to expand “Project Glasswing.”
Under that program, Anthropic planned to share a version of Mythos with select firms to help find and patch software vulnerabilities. Officials rejected the plan because of the risk the model could fall into the wrong hands and to protect the NSA’s offensive cyber capabilities. Bessent has led decisions about which foreign financial regulators and banks might access Mythos. In practice, Mythos is operating under a de facto temporary licensing regime.
Brad Carson, a former Oklahoma congressman who leads the advocacy group Americans for Responsible Innovation, called Mythos the AI-regulation fight’s “El Alamein moment.” El Alamein was the autumn 1942 battle that first proved British forces could defeat the Germans in World War II.
Churchill described that battle as “the end of the beginning.” Before El Alamein Britain had not truly prevailed; after it, the campaign turned. “Just as El Alamein wasn’t the end of the war, this fight isn’t over,” Carson said. “But the fall of Berlin is in sight. AI regulation will happen.”
Mythos’s influence extends beyond domestic policy — it has reshaped expectations for international AI governance. One notable outcome of last week’s Trump-Xi summit, The New York Times reported, was agreement to hold U.S.-China talks on AI safety. Until now, accelerationists who shaped Trump’s AI policy had strongly opposed any engagement with China, arguing such deals would constrain U.S. AI while China would not keep its commitments. They also used the “China card” to argue against domestic regulation.
Carson says that card has lost its potency. Most Americans worry about job losses from AI more than Chinese AI dominance. Mythos also appears to have convinced officials in both countries that allowing nonstate actors to gain dangerous cyber capabilities serves neither side’s interests.
Some in the administration now recognize that China-focused AI policies have not fully achieved their goals. Both the Trump and Biden administrations imposed export controls on AI chips and semiconductor equipment to prevent China from developing powerful military AI systems.
But Jacob Feldgoise, a senior data analyst at Georgetown’s Center for Security and Emerging Technology, says there is little evidence export controls have prevented China from acquiring strategic AI capabilities. Many military AI applications — autonomous drone navigation or target detection from satellite imagery, for example — do not rely on large language models that require massive fleets of high-performance GPUs. Even for LLM-based decision-support systems, Chinese tech firms trail U.S. AI labs by only about six months.
“Even if export controls didn’t stop China’s military AI development, GPU shortages likely slowed AI’s commercial spread across the economy,” Feldgoise said. That suggests the U.S. could consider easing some export controls in exchange for Chinese cooperation on an international AI governance framework. What that framework would look like remains an open question.
What’s clear is the tone in Washington has shifted dramatically. The city’s formerly hostile stance on AI regulation is crumbling.
/ Jeremy Kahn Darin Kim, reporter quill@fortunekorea.co.kr
Most Commented