Read the top story of the day in your inbox. Sign up for the Daily Prospect newsletter here.

Amid growing mistrust toward frontier AI labs and the data centers powering their models, AI industry groups have spent months pushing for a critical infrastructure designation that would expand the ability of both AI companies and federal law enforcement agencies to exchange information on U.S. citizens. The new designation would enshrine artificial intelligence in the same national security apparatus applied to dams, nuclear facilities, and hospitals.

The recent outages of critical software in aviation and banking, reportedly driven by hastily added patches to cybersecurity vulnerabilities identified by AI, may superficially lend credence to the idea that the data center locations where frontier models are produced need the highest level of protection. But the critical infrastructure designation, sources tell the Prospect, would serve largely to step up surveillance activities on individual critics of AI models, which based on the midterm election results includes a majority of the country.

The Prospect in your inbox

Analysis that goes beyond the noise with in-depth, progressive reporting.

No paywall, no programmatic ads — just the reporting. Unsubscribe anytime. Submitting your email signs you up for The Daily Prospect, our weekday morning newsletter.

In a 20-page report released in August, the AI advocacy organization Americans for Responsible Innovation argues that the technology’s integration into almost every other critical infrastructure sector means that it is now a new category unto itself. In order to address this integration, the group calls for an executive national security memorandum granting AI data centers special critical infrastructure status, instructing the intelligence community to “conduct intelligence estimates on AI sector risks and, as permitted by law and regulation, share those with industry” while also facilitating “public-private engagements across interdependent CI [critical infrastructure] sectors.”

Meanwhile, in an April hearing before the House Homeland Security Committee, retired Rear Adm. Mark Montgomery, a senior director at the Foundation for Defense of Democracies, told members of Congress that it was essential for elected officials to pass legislation securing AI data centers as critical infrastructure.

“If our nation does not properly secure these assets, our adversaries will steal, corrupt, and disrupt the data and communications that allow our economy to function,” Montgomery told the Cybersecurity Subcommittee.

As industry groups lobby, at least one AI company is explicitly furthering its integration into critical infrastructure sectors. At the beginning of October, Anthropic launched a Critical Infrastructure Defense Program to secure “the operational technology” behind power grids, water systems, and transportation networks, and protect government systems by bringing “frontier models, on-site engineers, and threat research to the defenders that protect operational technology.” In addition to integration into hospitals, power plants, and defense infrastructure, the case is now being made that AI is critical infrastructure for critical infrastructure.

According to two former counterterrorism officials who spoke to the Prospect on background, critical infrastructure designations open up increased surveillance by the Department of Homeland Security’s Office of Homeland Security Situational Awareness. They also open a spigot of federal DHS and Department of Energy grant funding for hardening infrastructure assets.

The Cybersecurity and Infrastructure Security Agency (CISA), also housed inside DHS, would be granted expanded authorities to monitor individuals under the new designation, while information sharing and prioritization inside the FBI would increase. As of today, CISA’s critical infrastructure designations have ballooned to include 16 sectors.

The designation would grant frontier labs more classified briefings, integration into critical infrastructure advisory councils, and increase intelligence sharing through both the FBI and DHS critical infrastructure information-sharing networks.

The proliferation of critical infrastructure safeguarding through public-private partnerships is not new, even if the technologies and systems it is now being applied to are. Energy companies were brought into meetings with the FBI to confront activists during the protests against the Keystone XL Pipeline in 2012, with the justification of safeguarding critical infrastructure. As Grist reported more than a decade later after wrangling documents through the Freedom of Information Act, the FBI held a summit with DHS and the oil company that owned the pipeline.

Members of DHS, the National Guard, state and local police, and the FBI attended the “regional working group meeting” in Oklahoma City to “ensure coordination and resource management between bureau field offices affected and the domain stakeholder, TransCanada Corporation.” A presentation from the Nebraska Information Analysis Center, one of 80 fusion centers connecting DHS and FBI with state and local police, showed that intelligence from CISA’s National Infrastructure Advisory Council created information intended to help charge Keystone protesters.

Because critical infrastructure intelligence-sharing platforms flow in both directions between the public and private sectors, vast numbers of critical infrastructure facilities and the corporations overseeing them are constantly exchanging real and perceived threats with federal law enforcement. This circulating information stream now includes warnings about a new threat dubbed “anti-technology extremism.”

The novel iteration includes a vast array of beliefs from those opposed to technology for religious, ecological, or social reasons. It has also been taken up by the State Department, where in July, Secretary of State Marco Rubio told U.S. allies that the U.S. is actively working to surveil and prosecute “anti-tech” extremists.

The lobbying push for a new designation comes as Donald Trump has signaled his unwavering support for a deregulated AI industry, instructing federal agencies to now refer to artificial intelligence as “Super Intelligence.” Meanwhile, private-sector AI boosters like Kevin O’Leary have called for investigations into activists organizing for democratic AI infrastructure development. That call was taken up by Republican members of Congress on the House Energy Committee, who wrote to the FBI to encourage an investigation into foreign influence.

The Prospect in your inbox

Analysis that goes beyond the noise with in-depth, progressive reporting.

No paywall, no programmatic ads — just the reporting. Unsubscribe anytime. Submitting your email signs you up for The Daily Prospect, our weekday morning newsletter.

But some of the facts about the dangers posed by activists appear to be dead wrong. In a June report compiled by the American Security Project and members of Columbia University’s School of International and Public Affairs, which bore many of the hallmarks of AI-generated text and formatting, protesters and activists are identified as an explicit justification for a critical infrastructure designation:

“Protesters and anti-AI groups have launched homemade explosives at external-facing data center components like power generators and energy support systems, with intelligence agencies warning of the potential for escalating future attacks,” the report claims. “While physical attacks on domestic data centers are more likely to be conducted by U.S. persons than state-sponsored actors, espionage and sabotage are common state-sponsored activities.”

The citation for the claim is an article published by me, describing federal overreach in surveilling AI opponents by alleging violent activity. The cited article does not provide any evidence of activists using homemade explosives against data centers. In short, it appears that an AI model hallucinated a threat against itself to justify surveillance of activists opposed to unregulated AI.

“This particular research project was conducted alongside a team of student researchers at Columbia University, but ASP holds ownership and responsibility for the accuracy and quality of the final product,” Matthew Wallin, CEO of the American Security Project wrote to the Prospect in an email. “This error indicates it is not up to our standard, and we are taking efforts to address this. We are also consulting with our colleagues at Columbia about the issue.” The report has since been taken offline, though an archived announcement of the report can be found.

Wallin added that “ASP maintains a ‘no AI generated writing’ policy for all of its written products and this is communicated to the entire project team. Whether this error was caused by unauthorized usage of AI, I am unable to verify at this time.”

Daniel Boguslaw is an investigative reporter based in Brooklyn.