Artificial intelligence now underpins major parts of the U.S. economy, from how financial institutions forecast and trade to how the energy industry plans and secures the power grid.
But should AI be considered “critical infrastructure” in the eyes of the federal government? Should it be legally designated in the same way as energy infrastructure, water systems and emergency services — sectors so important to U.S. society that disruptions could be disastrous?
The question is percolating among AI researchers — a community that has become increasingly influential as the fast-moving technology prompts novel safety questions — and even made its way to the acting director of the Cybersecurity and Infrastructure Security Agency, which oversees such designations.
As lawmakers grapple with doomsday warnings about AI and the federal government incorporates the technology into everything from military operations to air traffic control, some of these researchers see the potential for the idea to make it to the White House.
Trending
“As this technology becomes more pervasive and more important to everyday U.S. lives, and it becomes an increasing target for threat actors, there really needs to be front-end resilience built in,” said Jess Maksimov, one such researcher, a policy analyst for Americans for Responsible Innovation. “Disruptions can really be cascading and debilitating.”
Maskimov’s white paper, written with co-researcher Terrence Kelly, is among several recommendations making the rounds in Washington on ways to secure AI infrastructure. Another report, from the American Security Project and researchers at Columbia University, recommended designating data centers as a critical infrastructure sector, as well as increasing funding and personnel for CISA, which manages coordination around those sectors.
At a June conference, CISA acting director Nick Andersen was asked whether he considered AI to be “critical infrastructure.” Initially, Andersen said no.
“I think the things that go into it and make it work are probably critical infrastructure,” he said, “which maybe is a distinction without a difference.”
But he acknowledged that a sector designation could be possible. And at least, he said, formal coordination among frontier AI labs and the federal government will increase, in an attempt to answer three defining questions: “How do we secure from adversarial AI, how do we secure our AI, and how do we use AI to help secure ourselves?”
In a statement to The Washington Sun, Andersen said his agency recognizes the transformative power of AI, but he didn’t comment on whether he supported a sector designation.
“We are committed to promoting the secure use of AI while ensuring that innovation—crucial to America’s global competitiveness—is not hindered. We actively collaborate with our partners on shared priorities, staying informed about the latest advancements in AI and exploring how these technologies can strengthen the nation’s defenses against cyber threats,” he said.
Currently, the federal government has designated 16 sectors as critical infrastructure, including banks, dams and military supply chains.
Giving AI this designation would create a way for the industry to coordinate with federal agencies on security risks. It could also subject AI companies to reporting requirements for major cyberattacks. Under the direction of Congress, the Cybersecurity and Infrastructure Security Agency could also establish security standards for AI infrastructure.
The idea has gathered interest in both conservative and liberal circles, despite being largely absent from AI policy conversations in the White House or on the Hill.
Chris Krebs, who led CISA under the first Trump administration, said as digital systems have become a core part of modern conflict and run-of-the-mill crime, AI is enhancing threats. A “critical infrastructure” designation could help create a forum for regular conversations between AI companies and different parts of the federal government, like the Energy or Commerce departments.
A designation could streamline the way the industry shares information with federal officials and allow for AI companies to work together on safety standards without concerns about market manipulation or price fixing.
Krebs noted that under the Trump administration, the Department of Homeland Security has made changes to the framework for partnerships, and it remains to be seen what impacts that will have.
“Companies have been doing this stuff for a very long time, and things have, on the whole, gotten better as a result of these sorts of collaborations,” Krebs said. “I would rather have a mechanism like this than nothing at all, and possibly even [more than] really aggressive regulation.”
But there are a lot of details to work out. Among cybersecurity experts, it brings up questions around what counts as AI infrastructure, what powers CISA could exercise over the industry, and whether the embattled agency has capacity to take on a new sector.
The agency lost a third of its workforce after layoffs, buyouts, early retirement and reassignments of personnel to ICE and Customs and Border Protection.
Frank Cilluffo, director of the McCrary Institute for Cyber and Critical Infrastructure Security at Auburn University, said AI is “without a doubt” critical. But he said he was worried that a designation could function like a checkbox, letting lawmakers and federal officials off the hook for other kinds of safeguards. And, he said, he’d want to understand exactly what kinds of potential regulation the move would entail.
“A designation of critical infrastructure doesn’t have to lead to a regulatory approach, but that is the natural gravitation that some of the agencies would go to,” he said.
That potential for additional layers of red tape has sparked skepticism.
“I think it sounds nice to do, but I also think it doesn’t bring a whole lot of real safety and security. It brings a lot more bureaucratic ordination,” said Mark Dalton, a policy director at the center-right R Street Institute. “I don’t think actual tangible security benefits arise from that.”
There may be too many sectors already, he argued. “If 16 things are critical, then which one of those is critical?” he said.
President Donald Trump has called fears around AI safety a “hoax.” Even so, his administration recently proposed a shared notification system that would allow China and the United States to alert one another to AI incidents that threaten national security.
Proponents of making AI “critical infrastructure” say it won’t necessarily lead to more regulations.
“It’s a partnership, really,” Maksimov said. The public sector provides insight, resources and coordination on national security concerns, but “the onus is mostly on industry to make sure that their specific technologies are secure.”
No new critical infrastructure sectors have been designated since 2017, when then-Homeland Security Secretary Jeh Johnson created an election infrastructure subsector. The 16 sectors currently designated “critical infrastructure” were given that label under a directive from President Barack Obama.
AI increasingly seems to meet the definition of critical infrastructure, which the law says is “so vital to the United States that the incapacity or destruction of such systems and assets would have a debilitating impact on security, national economic security, national public health or safety, or any combination of those matters.”
The researchers say there are novel threats in the AI industry.
Some of those threats are popping up in other “critical infrastructure” sectors. Last month, federal officials warned of an “active threat” to manufacturing, energy, water, chemical, food and commercial facilities sectors from AI-powered hacking efforts targeting industrial control equipment.
AI infrastructure has also been a target for military action. During the U.S.-Iran war, Iranian drone strikes hit Amazon data centers in Bahrain and the United Arab Emirates, causing damage that affected banking operations and destroyed data.
Anthropic, OpenAI and Google did not respond to questions about whether they would support a critical infrastructure designation.
Designating AI as critical infrastructure under CISA would require action from the president, from Congress or both.
The White House did not respond to a request for comment. But Trump has already directed CISA to create a “cybersecurity clearinghouse” and coordinate research on software vulnerabilities and assessments of new AI models. And recently, he said he plans to appoint a czar to oversee an “AI Force.”
Stephanie Justice, a spokesperson for Americans for Responsible Innovation, said that could be a path toward a designation.
“Now that President Trump has discussed his plans for an AI Force, we urge him to use this new body as a way to secure AI as a critical infrastructure,” she said.
That level of interest hasn’t quite reached Capitol Hill.
“I haven’t given that thing any thought,” said Sen. Mark Kelly (D-Arizona) when asked if he would support such a move.
“I don’t know about that,” said Sen. Josh Hawley (R-Missouri). “That’s an interesting question.”
Author
