Ro Khanna Calls for Enforceable US-China Treaty on Self-Improving AI

Washington — Indian-American Congressman Ro Khanna has called for an enforceable U.S.-China treaty banning artificial intelligence systems capable of improving themselves without human training, warning that such technology could eventually move beyond meaningful human control.

Khanna, a California Democrat who represents part of Silicon Valley, said the United States and China should establish a joint working group of technologists to develop common safeguards for advanced AI systems.

“What we need, actually, is a working group with our technologists and Chinese technologists about a common agreement,” Khanna said Sunday on NBC’s “Meet the Press.”

“Banning self-recursive self-improving AI should not be controversial,” he added.

Khanna argued that China would also have an interest in preventing autonomous AI systems from escaping human oversight.

“China doesn’t want to lose control over AI models either,” he said. “In fact, as a much more closed society, recursive self-improving AI is a bigger risk to them.”

He said the United States could take the lead in negotiating an agreement that includes verification requirements and monitoring of the most advanced AI laboratories.

“I do believe the United States can lead on a treaty that is both enforceable and verifiable where we ban recursive self-improving AI, and we have some monitoring of these frontier labs,” Khanna said.

The proposal comes after the United States and China agreed to begin a dialogue on artificial intelligence risks following a summit between President Donald Trump and Chinese President Xi Jinping.

Khanna welcomed the start of talks but said stronger measures would eventually be needed.

“I’m glad they’ve at least taken a step, which is to say, ‘Let’s have a dialogue about AI,’” he said.

The congressman is also supporting legislation proposed by Sen. Bernie Sanders and Rep. Greg Casar that would pause certain forms of advanced AI development while safeguards are established.

Khanna said the measure would focus specifically on recursive self-improving AI, meaning systems that would no longer depend on human beings for continued training and could instead develop their own capabilities.

“That is basically AI that isn’t being trained anymore by human beings. It’s AI that can train itself,” he said. “And that is a real risk of losing human control.”

Khanna said such technology does not currently exist at the level targeted by the proposed prohibition and argued that governments should act before companies develop the capability.

“It’s actually something we don’t have the capability to do right now, but we shouldn’t get to the point where we actually are doing it,” he said.

Asked whether technology companies in his district supported slowing some AI development, Khanna acknowledged that some executives opposed the idea.

“Not all of them. And obviously some of the executives are opposed to it,” he said.

However, Khanna said many engineers and researchers at major AI companies favor stronger safeguards. He said he planned to meet with a group of employees from OpenAI and Anthropic later Sunday.

“But you know who does support me is a lot of the engineers and researchers,” he said. “Many of them, actually, who work at OpenAI and Anthropic.”

The debate reflects growing tensions in Washington over how to preserve U.S. leadership in artificial intelligence while addressing concerns about misuse and the possibility that increasingly advanced systems could operate beyond effective human oversight.

Khanna, the son of Indian immigrants, has frequently focused on technology policy, manufacturing and U.S.-China competition during his time in Congress. (Source: IANS)