Top executives at OpenAI, Anthropic and other leading artificial intelligence companies are quietly wargaming how to manage the public revolt and political fallout after a “catastrophic AI event,” according to Axios.

OpenAI says its scenarios are not treated as inevitable, but industry insiders have told Axios they expect such an event in the next six to 12 months.

The planning reportedly focuses on educating Congress and shaping legislation that could follow. But an equally urgent question is being neglected: If a revolt arrives, what happens to the physical infrastructure of artificial intelligence?

Data centers have quickly become foundational to the American economy and national security, as the U.S. races China for AI supremacy. These U.S. facilities are highly concentrated, which bolsters efficiency but also exacerbates fragility. Loudoun County, Va., hosts more than 250 data centers; on July 22, a fault on a single transmission line in Northern Virginia caused 3,800 megawatts of data center load to drop off, with effects felt as far as Chicago. If an accident can produce that, a deliberate attack could be far worse.

There are early indicators of an anti-AI revolt. In April, someone fired more than a dozen rounds into the home of an Indianapolis city councilor who had supported a local data center. The attacker left a note, reading “No data centers.”That same month, the home of OpenAI CEO Sam Altman was attacked twice, first with a Molotov cocktail and then with gunfire. Per the criminal complaint, the man charged in the firebombing cited AI’s risk of human extinction.

The AI industry is adding tension with its own warnings. In 2023, leaders of the top AI companies signed a statement comparing the risk of extinction from AI to nuclear war; later, Dario Amodei warned of mass unemployment. When the companies building AI describe it as catastrophically dangerous and economically displacing, some people will conclude they are obligated to stop it. A catastrophic event would enlarge that group overnight.

Attacks on infrastructure are an old extremist ambition. “The Turner Diaries,” the 1978 neo-Nazi novel that inspired the Oklahoma City bombing, depicts strikes on computer systems, power stations and utilities as a way to bring down the government. In 2021, the FBI arrested a Texas man who plotted to bomb an Amazon data center in Virginia in the hopes of crippling the internet, and in 2022, gunmen attacked two electrical substations in North Carolina, leaving tens of thousands in the dark for days.

While animus toward AI data centers is largely organic, adversaries are working to inflame tensions. In August, X reported that roughly 200 accounts connected to a Chinese bot farm were amplifying narratives encouraging Americans to oppose data centers. The New York Times reported that a Chinese state-owned newspaper published a satellite image of a data center in Virginia, writing that AI posed a threat to Americans’ physical and financial health. These efforts cost our adversaries very little, while adding fuel to a debate in which some have turned to violence.

Drones have also made physical attacks more accessible — and data centers are increasingly seen as a target. Earlier this year, Iranian drone strikes damaged Amazon Web Services data centers in the United Arab Emirates. This month, Ukrainian drones hit a major data center operated by Yandex, Russia’s leading AI company. Yandex said the attack forced it to fully suspend the center’s operations. And in research published in June, one of us documented Chinese vendors advertising drones rigged with drop munitions along with fiber-optic guidance systems that are immune to jamming, on Tiktok and Instagram.

An inexpensive hobbyist drone carrying a small charge can disable power and cooling systems that cost millions of dollars and take months to replace.

Growing public animus, adversaries working to inflame it and inexpensive drones that are easily weaponized are converging on facilities whose physical security falls in a gray area. Data centers are classed as part of the information technology sector, but the framework has not kept pace with their growing strategic importance and rapidly changing threat environment. Washington must respond on three fronts.

First, the president should direct the secretary of Homeland Security to establish a dedicated physical-security strategy for these facilities, with the Cybersecurity and Infrastructure Security Agency conducting a 90-day assessment of the threats they face and where they are most exposed.

Second, Congress should require the operators of these facilities to share information with federal authorities and local agencies. Word of a plot against a data center or a drone over a substation does no good if it stays inside a company’s security office.

Third, Congress should fund the authority it has already given to local law enforcement to defend against drone threats. The Safer Skies Act requires specialized certification and training through the FBI’s National Counter-UAS Training Center in Alabama, the only facility that can issue the credential, which has so far qualified only a small fraction of the country’s police departments. Congress should expand the certification pipeline and increase funding for detection and mitigation equipment.

The AI industry is right to prepare for the day after a catastrophe. Washington should prepare for the threats already emerging, starting by deciding who is accountable for protecting physical infrastructure where violent anger is likely to land.

Alex Goldenberg is a senior fellow at Rutgers University’s Miller Center for Community Protection and Resilience and the founder of Silent Index, a national security advisory firm. Robert Liscouski was the first assistant secretary for infrastructure protection at the Department of Homeland Security, where he built the offices that became the foundation of the Cybersecurity and Infrastructure Security Agency.