AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › People & Organizations

Gregory C. Allen

medium confidence · updated 2026-09-18

Director of the Wadhwani AI Center at CSIS; former DoD JAIC founding director of strategy and policy; leading US commentator on AI export controls and US-China AI competition.

Contents

Gregory C. Allen is a US national-security analyst who writes on AI export controls and US-China AI competition. He directs the Wadhwani AI Center at the Center for Strategic and International Studies (CSIS).

Background and roles

Allen directs the Wadhwani AI Center at CSIS. He was previously founding director of strategy and policy at the US Department of Defense Joint AI Center (JAIC), the predecessor of the Chief Digital and Artificial Intelligence Office (CDAO).

Positions and statements

Allen is among the most frequently cited US commentators on AI export controls and their effectiveness in constraining Chinese AI development.

Export controls after DeepSeek

In analysis published after the emergence of DeepSeek (CSIS, 2025), Allen reframed the export-control debate from a "controls failed" reading to the position that controls have design flaws and enforcement gaps while the core premise remains sound. He argues that the H20 episode showed Huawei's constraint migrating from silicon access to software ecosystem quality, which he describes as a more durable constraint.

The March 7, 2025 report identifies the design flaw as the October 2022 rules' treatment of the A800, which let roughly $9 billion of advanced chips reach China legally over a 12-month window, and the enforcement gaps as H100 smuggling networks (at least eight, each handling transactions above $100 million) and a Bureau of Industry and Security he describes as "drastically under-resourced," with fewer than 600 employees and a budget of about $200 million. He reports that TSMC manufactured more than 2 million Ascend 910B logic dies for Huawei through shell companies, which he calls a "strategically significant stockpile," and cites DeepSeek chief executive Liang Wenfeng's statement that export controls are "the greatest challenge" as the strongest direct evidence that the controls bind. His net judgment is that controls can "slow and disrupt, not stop" Chinese AI development, and his forecast is that the US lead is unlikely to exceed "more than a year or two, even with extremely aggressive export controls." He recommends restoring BIS and intelligence-community capacity to Cold War–era levels for adversary-technology controls, enforcing the January 2025 Foundry Rule, and closing logic, packaging and HBM chokepoints together (CSIS — DeepSeek, Huawei, Export Controls, and the Future of the U.S.-China AI Race (Allen, March 2025)). His TSMC die figure is more conservative than SemiAnalysis's estimate of roughly 2.9 million dies; see Chip Smuggling and Export-Control Evasion and AI Race Dynamics.

Washington's response to AI risk

In an October 2023 Stratechery interview about the updated China chip ban, Allen said that the Center for AI Safety's statement on extinction risk, signed by Sam Altman and other AI chief executives, had an impact on Washington that was "really difficult to overstate": "when you get that kind of roster saying, 'When you think of my technology, think of nuclear weapons,' you definitely get Washington's attention." Ben Thompson quoted the remark in arguing that such industry warnings should be read in light of the signatories' incentives (Attenuating Innovation (AI) — Ben Thompson (Stratechery, October 2023)).

Selected publications

Beyond the 2025 export-controls report, Allen's work is cited across other sources on AI and national security:

Relationships

Categories: People & Organizations | us | national-security | compute | government
What links here (3)