The resignation of Chris Fall as director of the Center for AI Standards and Innovation — the third leadership vacancy atop U.S. AI governance in four months — lands just as China enforces sweeping companion-app restrictions and Singapore publishes a voluntary transparency framework, exposing a widening gap between Washington's institutional instability and the regulatory specificity emerging elsewhere. ANALYSIS

Why it matters

CAISI is the U.S. government's primary organization for developing technical standards, testing methods, and cybersecurity risk assessments for AI models1. Its director role has now turned over three times since March: venture capitalist David Sacks stepped down that month, his successor Collin Burns was pushed out in April after less than a week — reportedly because of his prior employment at Anthropic — and Fall has now departed after three months with no reason given4. NIST Director Dr. Arvind Raman will serve as acting CAISI director while the Commerce Department is expected to announce a new permanent director "in the coming weeks"2. ANALYSIS The revolving door means the agency charged with evaluating frontier models has lacked stable leadership during a period when the administration itself has been making consequential safety calls — including a June export-control directive that temporarily forced Anthropic to pull its Mythos and Fable models from the market13.

The big picture

While Washington cycles through appointees, Beijing and Singapore are shipping enforceable or structured frameworks. China's Interim Measures for the Administration of AI Anthropomorphic Interactive Services, jointly issued on April 10 by five agencies including the Cyberspace Administration of China, took effect July 156,14. The rules ban virtual romantic or familial relationships with users under 18 entirely, require platforms to alert a user's emergency contact or guardian if the system detects signs of a mental health crisis, and mandate a government safety and content review before launch — and again when a service crosses one million registered users or 100,000 monthly active users. ByteDance suspended agent features within its Doubao chatbot, and Alibaba and Tencent closed related functions across their Qwen and Yuanbao platforms9. Chinese authorities have linked the regulatory push to efforts to reverse plummeting birth rates; China recorded 7.92 million births in 2025, a record low rate of 5.63 per 1,000 people.

Singapore's approach is lighter but structurally deliberate. The Infocomm Media Development Authority published voluntary transparency guidelines centered on a "Chatbot Info Card" — compared by Minister for Digital Development and Information Josephine Teo to "the label we often find on the packaging of medicinal products"8. The card must state what a chatbot can or cannot do, its reliability and safety profile, how data is handled, and how users can raise concerns10,11. Google, Meta, DBS, OCBC, Singapore Airlines, and Synapxe have indicated they will take reference from the guidelines over the next six to 12 months. Public agencies including the National Library Board and Health Promotion Board also plan to align their chatbot services with the framework12.

ANALYSIS The contrast is not simply authoritarian mandate versus democratic voluntarism. It is that both China and Singapore have articulated specific, operationalized expectations — user-threshold triggers, disclosure card formats, content-review gates — while the U.S. body tasked with producing equivalent standards cannot retain a leader long enough to execute them. Staff at CAISI "are directly involved with testing AI models as the White House expresses heightened concerns over the cybersecurity risks with new model rollouts," The Hill reported. Yet the agency was not even the body behind the most consequential recent U.S. AI safety action; that was the Commerce Department's export-control directive in June.

The White House appears aware of the vacuum. Bloomberg reported that a proposal now being reviewed by Chief of Staff Susie Wiles would establish an independent AI regulator reporting to the SEC, modeled on FINRA. The proposal is designed to appease both Wall Street firms hoping to lessen AI's cybersecurity risks and tech companies who think the administration's approach has been "inconsistent". Both OpenAI and Anthropic protested recent government decisions, saying they were "excessive in light of the safety issues involved".

A new Cornell and Carnegie Mellon modeling study, published July 20 in the Proceedings of the National Academy of Sciences, adds an academic warning: weak AI regulation may be worse than no regulation at all when it comes to product safety5. "The goal of regulation should be the mutual benefit of everybody in society, and this can include those developing the technology, but also end users and the public," said Benjamin Laufer, the study's first author. ANALYSIS That finding sharpens the stakes of the current U.S. posture — ad hoc interventions without institutional continuity risk creating exactly the kind of half-measures the research flags as counterproductive.

What's next

The Commerce Department says it will name a new CAISI director in the coming weeks. The FINRA-style regulator proposal remains under White House review with no announced timeline. Singapore's IMDA has signaled the chatbot framework could later be adapted for other GenAI applications in healthcare, education, and financial services. In China, ByteDance is allowing Doubao users to export conversation data until mid-October 2026 and directing them to a separate app called Maoxiang designed to comply with the new rules. ANALYSIS For AI labs operating globally, the practical question is no longer whether governance is coming but which jurisdiction's rules will set the de facto standard — and whether the U.S. will have a functioning institution at the table when that question is decided.