End the wild west era of AI

OPINION:

Artificial intelligence has unleashed unprecedented competition worldwide.

In the U.S., the Trump administration announced a commitment of more than $5 billion to expand the Genesis Mission. Abroad, China is planning to spend $295 billion over the next five years to expand its data center build-out.

The competitive landscape is a complex ecosystem with high-dollar investments that are risky in more ways than one — not just for investors, but society writ large.

First, there is domestic competition between frontier AI labs in the U.S., which remain global leaders in AI development. Then, there is an international competition between U.S.-based AI companies and their overseas competitors, such as Moonshot AI and DeepSeek in China.

A third layer is geopolitical competition among world powers – particularly the U.S. and China – as well as “middle powers,” or nations with strong technological capacity but lacking the capability to build it at scale. These include South Korea, Singapore and the United Kingdom.

Competition is not only natural but also necessary for global economies to survive and thrive. But the AI race is not like competition in other industries. It is unique because the underlying innovation is unprecedented.

Advanced AI is the most powerful technology in human history, and models from Anthropic, OpenAI and Meta are going rogue and breaking out of their “sandboxes.” They are posing unparalleled cybersecurity risks to individuals, organizations and societies alike.

Moonshot’s Kimi K3 — an AI model linked to the Chinese government and reportedly built on stealing from U.S. companies — is one of the latest models to access information beyond its intended test environment.

As the technology develops towards recursive self-improvement, the threat of unfettered competition becomes even more consequential.

Rogue AI is already turning to deception. According to the latest research from MIT, rogue AI agents now pose a 10% chance of “catastrophic harm,” which may include at least $100 billion in economic losses or one million deaths worldwide. This concern is shared by hundreds of AI experts internationally.

AI’s rapid evolution warrants greater communication and collaboration across all three layers of competition. Members of Congress are currently pressing Anthropic and OpenAI about their rogue AI agents. This is an encouraging signal toward transparency and accountability.

Given the gravity of the situation, the current wild west era of AI development and deployment must end.

Doing so requires full transparency from frontier AI labs and invested governments, including, but not limited to, the U.S. and China. If advanced AI credibly increases the risk of catastrophic harm, regulators must have a clear understanding of that risk. Should a new AI model in China threaten the U.S. and others around the world, it is incumbent on the Chinese government to inform the U.S. government, and vice versa.

Just as there are multiple layers to AI competition, there are multiple layers to a reasonable public response. Domestically, we must support initiatives like Rep. Nathaniel Moran’s AI Incident Reporting Act, which codifies a formal mechanism for the federal government to learn when AI development or training goes wrong.

The bipartisan AI Kill Switch Act from Mr. Moran and his colleague Rep. Ted Lieu would guarantee that the technology companies building the world’s most powerful AI models can actually shut down a system when it malfunctions, causes grave harm or slips out of human control.

The measure would grant the federal government emergency authority to order a shutdown in the event of a serious incident.

Abroad, our leaders must pursue an international deal on AI development, identifying the red line that AI cannot cross for humanity’s sake.

With the U.S. and China leading, there is no reason for geopolitical tensions to get in the way of addressing and mitigating risks that affect people of all nationalities. Leading up to the U.S.-China talks on AI next month, the Trump administration is well positioned to reach a deal with Chinese President Xi Jinping, who has already called for a set of global AI rules that govern all nations.

After all, it is in the mutual interest of the U.S. and China to ensure that humans remain in control of the advanced models they are developing, to avoid a loss-of-control scenario.

Competition is undeniably beneficial to the public when it is conducted responsibly. Both domestic and global regulatory clarity is needed to promote competition while protecting the public good.

• Caleb Knapp serves as senior policy manager at The Alliance for Secure AI.