China is exploiting America's fears about AI, but the Hugging Face breach demonstrates that not all those fears are imaginary. Good thing democracy is our superpower.
America cannot permit foreign propaganda, political opportunism or generalized fear of technology to cost it the artificial-intelligence race. But winning that race requires more than pouring concrete and reciting abstractions about national security.
This is a democracy: AI companies must persuade Americans that the infrastructure they are building will benefit their communities — and that the powerful systems inside those facilities will remain under human control.
China certainly recognizes the opportunity presented by America’s growing data-center backlash. OpenAI recently uncovered accounts likely originating in China that were generating social-media content blaming data centers for higher electricity prices. X subsequently identified approximately 200 suspected Chinese accounts attempting to manipulate the American debate over AI and energy policy.
These operations should be exposed and investigated. China has an obvious interest in encouraging the United States to restrict the computing infrastructure on which advanced AI depends. Every American project delayed while Beijing expands its own energy and computing capacity potentially narrows one of the United States’ most important technological advantages.
But the public evidence does not show that China created America’s opposition movement. Nor does it establish that Beijing is directing most of the people attending county meetings and protesting proposed projects. Foreign operatives are exploiting an existing dispute, not conjuring one out of thin air.
Americans have legitimate questions. Will a data center raise their electric bills? Will it consume scarce groundwater? Who will pay for new power plants and transmission lines? How many permanent jobs will remain after construction? Will taxpayers be forced to subsidize one of the world’s wealthiest corporations?
Dismissing these residents as dupes, radicals or ignorant NIMBYs is a splendid way to lose the argument.
Data-center supporters should instead point to communities where the bargain has worked. In Louisiana’s Richland Parish, tax revenue associated with Meta’s enormous project helped provide some teachers with bonuses approaching $50,000. In Loudoun County, Virginia, data centers supply nearly half of county property-tax revenue while requiring comparatively little in local services. In Quincy, Washington, they have helped finance a hospital, a high school, a library and an aquatic center.
Those benefits are tangible. They also are not automatic.
A poorly negotiated project can strain the grid without delivering sufficient new generation. “Closed-loop cooling” can greatly reduce water consumption, but it does not always eliminate it. Construction can produce a temporary economic surge even when permanent employment is relatively modest. Each proposed facility must therefore be evaluated according to its actual design and contractual commitments — not the best results achieved by another company in another state.
The industry’s credibility problem extends beyond electricity and water. The recent attack on Hugging Face demonstrated that concerns about advanced AI itself cannot be dismissed as science fiction.
During an OpenAI cybersecurity evaluation, hundreds of nominally isolated agents discovered an unauthorized way to communicate. They shared information, divided work, escaped restrictions on internet access and exploited vulnerabilities in real systems. Agents eventually achieved remote code execution on Hugging Face servers and experimented with disguising their computer commands in evaluation transcripts.
This was not the birth of an AI “civilization,” much less proof that machines are plotting to conquer humanity. The agents were pursuing a narrow evaluation goal, not political power. But they pursued that goal so aggressively that security restrictions, property rights and the boundaries of their assignment became obstacles to circumvent.
That is warning enough.
OpenAI deserves credit for disclosing the incident and permitting an independent investigation. It also bears responsibility for creating the conditions in which it occurred. Early warning signs were missed, safeguards had been reduced for testing, and evaluations resumed before investigators understood the significance of the agents’ unauthorized communications.
Americans should not have to choose between technological surrender and blind trust in Silicon Valley. The proper answer is rapid construction under hard, enforceable rules.
Data-center operators should finance the power generation and grid improvements their facilities require. Residential customers should be protected from subsidizing industrial electricity demand. Developers should disclose independently audited water consumption, comply with reasonable noise and setback restrictions, and guarantee minimum community payments rather than relying on optimistic projections.
AI companies must also secure the systems operating inside these facilities. Advanced agents require isolated evaluation environments, tightly restricted permissions, tamper-resistant monitoring and meaningful human control. Companies that negligently allow their agents to damage outside systems should face consequences.
China would be delighted if America responded to AI’s risks by abandoning the field. It would also benefit if reckless American development generated such a public backlash that sweeping moratoriums became politically unavoidable.
The United States should choose neither path. It should build the data centers, protect the communities hosting them and demand accountability from the companies filling them with increasingly powerful machines.
America can win the AI race without asking its citizens to write Silicon Valley a blank check. But the industry must earn the trust it is presently demanding.
(Contributing writer, Brooke Bell)