A prepared statement dated September 2, 2026 by David Feith, a senior fellow at the Hudson Institute and former Deputy Assistant Secretary of State for East Asian and Pacific Affairs, submitted to the US House Permanent Select Committee on Intelligence for its hearing Persistent Competition and Legacy Architectures. It is addressed to Chairman Crawford and Ranking Member Himes. The statement's argument is that "the United States needs a new counterintelligence map for the AI age," because an increasing share of strategically important technology now sits in private companies and commercial infrastructure rather than in government laboratories and a bounded defense-industrial base. It identifies five ecosystems requiring counterintelligence attention and closes with five recommendations.
The statement is a policy position submitted as testimony, not a finding of fact, and several of its central claims are themselves attributed by Feith to third parties.
Framing: the "Sputnik problem"
Feith opens with a composite Taiwan-crisis scenario — pre-positioned Chinese attack drones near US bomber bases, compromised Chinese-origin equipment inside data centers shifting electrical load to destabilize the grid across 20 states, hackers using AI trained on smuggled and remotely accessed US compute to blind missile defenses, and a frontier model stolen from a US lab — and states that "None of these scenarios is science fiction. We have already seen pieces of each one."
The section titled "The Sputnik Problem" argues that the phrase "Sputnik moment" is usually misapplied to an adversary's breakthrough, when what made Sputnik historically important was the American mobilization that followed. By that standard, Feith writes, "much of the China challenge has produced Sputnik warnings without a Sputnik response." He couples this with an argument that public skepticism of government competence, while deserving of respect, "cannot allow justified skepticism to become strategic paralysis."
The five ecosystems
Sensitive sites and drone exposure
Feith cites Ukraine's Operation Spiderweb of June 1, 2025, in which Ukraine trucked some 117 small drones deep into Russia to attack strategic bombers at four airfields, as a demonstration that inexpensive systems pre-positioned inside an adversary's territory can produce strategic effects without penetrating defenses from abroad. He notes Israel's reported use of a related model against targets inside Iran weeks later, and Iran's own use of drones across the Middle East.
Domestic incidents cited: unidentified drones over Virginia's Langley Air Force Base and nearby facilities for roughly 17 nights in December 2023; unauthorized drone incursions in March 2026 at Louisiana's Barksdale Air Force Base, home to B-52 bombers and Air Force Global Strike Command, which he says remain under federal investigation; and a trailer park adjacent to Missouri's Whiteman Air Force Base, home of the B-2, purchased through a web of companies whose foreign owners press reporting and members of Congress linked to individuals with alleged Chinese-intelligence ties, after which Missouri initiated cancellation proceedings against the operating partnership for state corporate-registration violations. He also notes that a Chinese national pleaded guilty in 2026 after repeatedly photographing Whiteman's perimeter and equipment.
On authorities, the statement credits Congress with giving CFIUS power to review certain foreign real-estate transactions near sensitive military installations and Treasury with expanding the covered-installation list and geographic reach in 2024, while stating that implementation of the February 2025 America First Investment Policy memo "has been slow and limited."
The power grid and data-center hardware
Feith argues that data centers create a grid-security problem that is new in kind rather than degree, because a campus drawing 300 MW to a gigawatt or more is "densely computerized and can change its electricity consumption almost instantaneously," unlike traditional industrial load.
The anchoring incident is a routine transmission fault in Northern Virginia on July 22, 2026, which he says caused roughly 3,800 MW of data-center load to transfer rapidly from the grid to backup generation, producing large voltage and frequency swings across the eastern United States from Virginia to Chicago. He states explicitly that operators stabilized the system, there was no blackout, and it was not a cyberattack — but that it "demonstrated the physics an attacker would study." The inferred threat is an adversary with synchronized access to enough inverters, batteries, uninterruptible power systems, or other energy controls manipulating large blocks of demand.
Cryptocurrency mines are presented as an existing instance of the same problem: Feith states that almost all US crypto mines rely on Chinese-made compute and equipment and many have Chinese ownership and control. He cites the 2024 divestment, ordered by President Biden, of a PRC-owned crypto-mining operation within a mile of Wyoming's Warren Air Force Base after CFIUS found its foreign-sourced equipment could facilitate surveillance or espionage against a node of the nuclear deterrent, and says the Senate Intelligence Committee has since warned about remotely controllable Chinese-made mining machines.
Two supporting claims are attributed rather than asserted: that the Chinese inverter manufacturer Deye demonstrated in 2024 that it could remotely disable installed equipment abroad during a commercial dispute; and that the research firm Strider has identified more than 2,700 Chinese research papers studying grid vulnerabilities, including hundreds it assesses as highly relevant to potential US grid attacks and research involving PLA institutions.
Feith uses Foxconn to argue that corporate nationality is not a sufficient security test — the company is Taiwanese and "a critical partner to American AI companies," but its role in AI-server and rack manufacturing alongside a longstanding Greater China footprint "illustrates why a company's flag alone cannot answer the supply-chain-security question."
The section states that this "is emphatically not an argument against building data centers," which Feith says America needs to build "faster and at extraordinary scale," and credits the AI Action Plan with getting the principle right that the AI stack and its supporting energy and telecommunications infrastructure should be free of foreign-adversary technology. It names the FCC's action on foreign power inverters and an executive order restricting foreign-adversary hardware in the bulk-power system as starts, and asks for visibility into the installed base, trusted-supplier and firmware standards, elimination of adversary remote access, network segmentation, and grid rules designed for gigawatt-scale digital loads.
Frontier AI labs
Feith describes frontier labs as private companies holding "strategic national assets," and contrasts them with the Manhattan Project: today's labs are not isolated, since "employees travel, change jobs, publish research, work with contractors and cloud providers, and operate enormous commercial networks," which drives innovation while creating an attack surface. He organizes the threat into three parts.
- Cyber. Labs face nation-state adversaries "with capabilities and intelligence that no private company can match." He asks for durable channels for classified threat sharing, incident coordination, defensive assistance, and government vetting and investigative support, with labs providing defense in depth across model weights, internal systems, data centers, cloud providers, and contractors.
- Insider threats. The January 2026 conviction of former Google engineer Leon Ding for economic espionage and trade-secret theft, after stealing more than 2,000 pages of confidential AI information while secretly building a business in China, "should be treated as a warning, not merely a successful prosecution" (United States v. Ding). Feith frames the open question as what government and industry learned and what changed, and asks that security address privileged access, conduct, undisclosed affiliations and anomalous activity "while protecting civil liberties and openness."
- Adversarial distillation. He cites the White House statement of April 2026 that foreign entities, principally in China, were conducting industrial-scale campaigns to distill US frontier systems using tens of thousands of proxy accounts, and argues that with enough access to high-value outputs "an adversary may reproduce strategically significant capabilities without ever stealing model weights." His counter-distillation list is identity controls, monitoring and rate limits, detection of coordinated activity through resellers and third-party routers, rapid revocation, intelligence sharing, and consequences for state-backed circumvention (Adversarial Distillation).
The section invokes NSA Director Keith Alexander's characterization, fourteen years earlier, of Chinese cyber-enabled IP theft as "the greatest transfer of wealth in history," and states that the value now at risk includes "the returns on hundreds of billions of dollars of American investment and potentially decisive strategic capabilities." An unnamed "astute commentator" is quoted for the claim that Chinese distillation "threatens to bankrupt the U.S. economy."
Feith also argues the converse: that frontier models are themselves defensive instruments that can help harden the grid, hospitals, water systems, and communications networks "before advanced offensive capabilities diffuse more broadly."
Advanced compute and the semiconductor supply chain
Feith calls compute "China's principal chokepoint" while granting China "world-class engineers, enormous energy resources, sophisticated AI companies, and a government willing to mobilize national resources." Two quantitative claims support the chokepoint reading: that America's leading AI chips are "roughly five times more powerful than China's best," with public roadmaps suggesting the gap could widen; and, attributed to the Institute for Progress, that Huawei may produce "only about one to two percent as much advanced AI computing capacity as the United States this year."
He argues the advantage "is not self-executing" and exists largely because US and allied export controls denied China both the best chips and the manufacturing equipment to make equivalents at scale. The organizing frame is capability rather than transaction: China "should not be able to buy, rent, smuggle, steal, or build its way around American technology controls."
| Vector | What Feith argues it requires |
|---|---|
| Buy | Controlling direct sales of the most advanced AI chips |
| Rent | Closing the remote-access loophole — a company barred from receiving a chip should not be able to rent thousands of identical chips in an overseas data center |
| Smuggle | Treating diversion as an industrial-scale enforcement problem across distributors, shell companies, freight forwarders and Southeast Asian transshipment hubs, following the networks rather than stopping at the first consignee |
| Steal | Recognizing TSMC, ASML and American chip companies as counterintelligence targets for people, processes, designs and manufacturing knowledge |
| Build | Restricting semiconductor-manufacturing equipment — lithography, deposition, etch, metrology, packaging, servicing and know-how — since controlling a finished GPU accomplishes less if China can make its own |
He concludes the section by arguing that "export control becomes counterintelligence, and counterintelligence becomes economic security," and that "the bureaucratic label matters less than the outcome" (Export Controls (AI), Chip Smuggling and Export-Control Evasion).
Startups and the counterintelligence threshold
The final ecosystem is defined by the difficulty of recognizing it. Feith argues that an extraordinary share of US innovation now occurs in startups working on space, quantum, fusion, biotechnology, advanced materials, autonomy, robotics, energy and manufacturing, and that such a company "can move remarkably quickly from unimportant to promising to actually important" — a prototype works, a material becomes essential to missile production, a space startup becomes central to military communications. At that point, he writes, the company has crossed what "we might call a counterintelligence threshold." The framing is drawn from a scene in the film Oppenheimer in which Ernest Lawrence tells Oppenheimer that the government cares about his associations "Because you're not just self-important, you're actually important."
The stated problem is that "Washington has no systematic mechanism for recognizing when that happens," so recognition typically waits on a defense contract, a chance government relationship, an intrusion, or a discovered insider problem. His proposal is a "graduated private-sector counterintelligence partnership" that stops short of turning every startup into a classified defense contractor: threat intelligence, designated counterintelligence contacts, voluntary security assessments, cyber assistance, insider-threat guidance and help navigating foreign investment, with stronger obligations attaching later where a company receives classified information or sensitive contracts.
Recommendations
The statement closes with five named priorities:
- Build a Technology CI Mission — an Intelligence Community–led effort with the FBI, Commerce, Defense, Energy and Homeland Security to identify private companies and infrastructure crossing the counterintelligence threshold and characterize the Chinese tactics targeting them.
- Protect Frontier AI — a durable government-industry security compact covering nation-state cyber threats, insider risks, model security and industrial-scale distillation, with counterintelligence "built into any emerging federal framework for frontier AI."
- Harden the Homeland Perimeter — extending CFIUS, land-security, base-defense, counter-drone, grid and hardware-security authorities to a world in which nearby commercial property can become a drone-launch site and remotely accessible equipment inside a gigawatt data center can become a grid vulnerability, and treating large data centers as critical grid infrastructure.
- Protect the U.S. Compute Lead — technology controls organized around capabilities rather than individual exports, with export enforcement, counterintelligence and allied controls treated as one mission.
- Reindustrialize What Matters — on the argument that "we cannot counterintelligence our way out of industrial dependence," extending the strategic logic behind the CHIPS Act to other supply chains, since restrictions on Chinese drones, inverters and transformers "will fail if American and allied alternatives do not exist at scale."
The conclusion states that the United States "does not lack warnings" and asks whether it will mobilize, while explicitly declining a fatalistic reading: Feith closes on US advantages in frontier labs, advanced compute, semiconductor design, capital markets, allies and talent attraction, arguing that "Counterintelligence is urgent precisely because we have so much to protect."
Reception
Trade coverage of the hearing on September 3, 2026 reported the testimony narrowly, as a call to treat data centers as a major cybersecurity target for adversaries including China (Source: insideaipolicy.com). Data centers are one of five ecosystems in the statement and occupy its third section.
Provenance
Captured September 4, 2026 from the Hudson Institute's own media host as a ten-page PDF with an intact text layer. The document is captioned September 2, 2026; the hearing was held September 3, 2026 per trade coverage, and the two dates should not be collapsed. The statement self-identifies the committee, the hearing title, and the chairman and ranking member addressed.
Relationships
- supports: AI Data Centers — grid-security and hardware-provenance framing for data centers
- supports: Adversarial Distillation — counter-distillation as a counterintelligence mission
- supports: Export Controls (AI) — the buy / rent / smuggle / steal / build framing
- supports: Chip Smuggling and Export-Control Evasion — diversion as an industrial-scale enforcement problem
- related: United States v. Ding — the Ding conviction as an insider-threat precedent
- related: US-China AI Competition: Different Races, Different Metrics
- related: AI and Cybersecurity, Compute Governance
- related: America's AI Action Plan — credited for the foreign-adversary-technology principle
- instance-of: AI and National Security