Ezra Klein Urges Washington to Ban Self-Improving AI Until It Is Proven Safe

Ezra Klein wants Washington to ban artificial intelligence systems from building more advanced versions of themselves until the practice is shown to be safe, a position at odds with President Donald Trump's stance that the United States is "not going to be putting on brakes" on AI. Klein made the case in a recent episode of The Ezra Klein Show.

His guest was David Robinson, who recently resigned from OpenAI after leading the writing of the company's safety system cards. Those are the technical reports that explain why a new model is considered safe to release. Robinson is a Rhodes Scholar and Yale Law School graduate who co-founded the civil rights nonprofit Upturn and advised the Biden White House. He told Klein the safety problem runs through the whole AI industry, not just OpenAI.

Ezra Klein's Proposed Ban on Recursive AI Self-Improvement

Klein summarized an argument from one of his own recent pieces: a default ban, with exemptions granted one at a time.

"Let's not do recursive self-improvement until we're sure it's safe. Let's just ban it and begin to carve out exceptions as we know the exceptions are safe," Klein said.

Recursive self-improvement, often shortened to RSI, means AI systems doing the research that produces their own successors. Robinson called it his main worry: "the AI can make a smarter AI in some way that we're not going to understand."

The industry's answer has been to slow down, which companies call "pacing the frontier." Robinson said, "I hate that term." Klein said slowing down is not enough: "Running off a cliff and walking slowly off a cliff are just not that different."

Robinson said more effort alone will not fix the problem. "I don't think that alignment is an engineering problem. I think it's a science problem," he said. Describing what he saw inside OpenAI, he said he looked around and thought, "this is nuts what's happening."

The China Objection Klein Raised

Klein then said he would "steel man this case" because he hears it "from people all the time." One critic, he said, asked whether Americans could still use "a Chinese open weights model that has been improved using recursive techniques." A second problem follows from the first: "if we slow down, China speeds up."

"I definitely don't want to see AI become a reason that China dominates the United States," Robinson replied. But he said politics can move quickly: "I definitely believe that that could happen with respect to US-China cooperation on AI."

The record shows some movement, though so far it is limited. At their Washington summit in September, President Donald Trump and Chinese President Xi Jinping agreed to set up a channel for AI-related incidents and to hold an AI dialogue in November. Trump later told reporters outside the White House: "The United States of America is not going to be putting on brakes."

Washington Has Moved Toward Voluntary Rules

The administration has worked against state AI rules. A December 11, 2025, executive order created a Justice Department task force to challenge state AI laws and tied some broadband funding to how states regulate AI. The White House said state legislatures had introduced more than 1,000 AI bills. Robinson recalled that by the time he moved to safety work, "there were hundreds of state bills to keep track of."

CNN's Fareed Zakaria has warned that the White House's AI safety rules are only voluntary.

Klein, co-author of Abundance, raised his own caveat. "We so aggressively regulated nuclear power, in my view, over-regulated nuclear power, that nuclear power broadly stopped being built in this country," he said. Robinson agreed, then added: "But I would much rather have those problems than the ones we do now."

What Polling Shows on AI Regulation

The public leans toward caution. A Gallup survey conducted April 25 to May 5, 2025 found that 80 percent of Americans want the government to keep AI safety rules even if that slows development. Only 9 percent put speed first. Support was 88 percent among Democrats and 79 percent among Republicans and independents. Just 16 percent said each company should set its own rules.

What Happens Next

U.S. and Chinese officials are set to hold an AI dialogue in November.

The question Klein's critic posed is still open: whether a U.S. ban could cover models built in China.