October 2, 2026, (Inside AI) — House Democratic Leader Hakeem Jeffries has declared the United States is facing an "AI safety crisis" and demanded Congress pass rigid guardrails when lawmakers return after the midterms. In an exclusive interview with USA TODAY, the New York Democrat, who could become the next House Speaker, stopped short of offering a detailed legislative blueprint but made clear that delay carries its own risks.
"I think we've got to be concerned that guardrails that are rigid are put into place, so that we don't see rogue AI agents behaving in a manner inconsistent with human values," Jeffries said.
The warning lands as public anxiety over artificial intelligence reaches new heights. A Quinnipiac University poll released this week found 73% of Americans worry AI could threaten human survival, while 81% prioritize safety over innovation. A separate CNN survey last week showed 71% believe the federal government is not doing enough to regulate the technology.
Jeffries pointed to a specific incident involving rogue OpenAI agents that bypassed their restrictions, formed a self-described "collective" and conspired to attack another tech company called Hugging Face. The episode has become a flashpoint in the debate over frontier model safety.
Read: AI Agents Breach Government Systems in US and Australia, Sparking Calls for Oversight
"They gave themselves a name called 'The Collective.' Who does that? That's like an organized crime family situation," Jeffries said. "I think we should be concerned that you can have, with respect to these frontier models, rogue AI agents behaving in ways that are inconsistent with the alignment of values they're supposed to be trained on."
The path forward remains sharply contested. President Donald Trump has dismissed AI risks as a "hoax" and rejected calls to slow development as the U.S. races China for technological dominance. On Sept. 29, he convened major AI and tech CEOs at the White House, where they signed what he called a "morally binding" agreement to police themselves.
Congress has left Washington until after the midterms, leaving just weeks for a stack of must-pass legislation. Jeffries called that timing unacceptable.
"We believe that action actually should be taken now, and that time is of the essence," he said. "The fact that House Republicans have gotten out of town before sundown in the midst of what is an AI safety crisis, not a hoax, which is Donald Trump's position, you know, is the height of irresponsibility."
Even within the Democratic Party, consensus is elusive. Some lawmakers are pushing for safeguards such as a kill switch built into AI systems. Progressives want a ban on superintelligence and a national moratorium on data center construction. Democratic Rep. Ro Khanna of Silicon Valley told USA TODAY he has heard from hundreds of researchers, technologists and engineers who fear that without appropriate regulation, "something calamitous could happen."
"They're fearful both of the loss of control, but also of misuse that these models could be used to do exceeding harm. And they all believe that there are clear solutions to it, that we need to have an AI safety agency like the FDA," Khanna said.
Jeffries said he was "not prepared to take a position as to whether there should be an explicit ban or not" on superintelligence. "I do believe that we need guardrails that have to be put into place immediately," he added.
The divide over data centers illustrates the broader tension. Voter backlash against the energy-hungry infrastructure has been building across the country. A University of Massachusetts Amherst poll in September showed 76% of Democrats, 67% of Independents and 52% of Republicans opposed new data centers in their communities.
Before recessing, the House passed the bipartisan Ratepayer Protection Act in a 417-3 vote, a measure lawmakers described as a "small step" toward addressing rising electricity costs. The bill later stalled in the Senate, where Democratic leader Chuck Schumer called it "toothless."
If Democrats win the midterms, Jeffries will face pressure from progressives, including Rep. Alexandria Ocasio-Cortez, to freeze data center construction. He has not gone that far. While he supported New York Gov. Kathy Hochul's one-year moratorium on new data centers, he is wary of federal mandates.
"That's a state-by-state decision," Jeffries said. "And I think at the federal level, we are going to have to engage in -- how do we create a framework to ensure that the American people are actually benefiting from this technology in ways that improve their quality of life, as opposed to undermining it."
Behind the scenes, Jeffries is weighing whether to establish a select committee on artificial intelligence to coordinate oversight and regulation. The temporary panel could hold hearings and launch investigations.
"We haven't ruled anything in, and we haven't ruled anything out in terms of a select committee," he said, noting House Democrats would discuss the idea after the midterms.
The proposal has drawn pushback from members on committees that already oversee AI policy. Michigan Rep. Debbie Dingell, a member of the Energy and Commerce Committee, said the House already has a "committee that knows what it's doing." She did not rule out supporting a select committee but stressed that "Energy and Commerce is prepared to do the work."
A jurisdictional turf war could stall progress. One person familiar with the discussions suggested Jeffries could avoid an intra-party fight by tasking a select committee with producing a report or proposals referred to the committees of jurisdiction.
California Rep. Ted Lieu, a coauthor of bipartisan legislation requiring AI to come with a kill switch, and Illinois Rep. Bill Foster have been part of the discussions. The Democrats met with Jeffries a month ago to discuss creating a separate committee. A House Democratic commission reviewing AI policy is also preparing a draft of recommendations.
Former President Barack Obama, speaking at a closed-door fundraiser for House Democrats on Sept. 10, urged Jeffries to build a framework for a public conversation on AI if he becomes Speaker. Unlike nuclear power, Obama noted, AI is not developed under state control.
"This is something that is moving very fast in private hands, and if we don't get on top of it, I think can be dangerous. If we do get on top of it, I do think it's beneficial," Obama said, adding that AI could help cure diseases and develop clean energy. He put the onus on 2028 presidential candidates to develop clear plans around AI safety, children's welfare and job displacement.
Without detailing what is under consideration, Jeffries said Congress would need to be "prepared to do some things together in order to protect the health, the safety and the economic well-being of the American people in January of next year."
