Author: Nita Farahany (with guest lecturer Chase Hamilton, Akin LLP) Source: https://nitafarahany.substack.com/p/when-silicon-valleys-existential Published: November 18, 2025
The 23rd class in Nita Farahany's AI Law & Policy course is a deep dive on US export controls, anchored by a guest lecture from Chase Hamilton, an export-controls attorney at Akin LLP. The class opens with a p(doom) check keyed to an Eliezer Yudkowsky TED talk — the class is split between under and over 50%, with none below 20% — and proceeds to frame export controls as a tool for effective-altruism-style existential risk management: controlling who builds AGI, under what governance, and over what timeframe.
Effective-altruism framing
Hamilton's lecture presents export controls through an effective-altruism lens. Beyond p(doom), the framing adds p(win) (described as unprecedented human flourishing) and p(utopia). The argument is that if democratic nations with strong safety practices are more likely to develop beneficial AI, export controls become instruments to improve humanity's odds.
ITAR versus EAR
The class distinguishes the two principal US export-control regimes. The International Traffic in Arms Regulations (ITAR), administered by the State Department, govern defense articles on the US Munitions List and operate on the principle that everything is prohibited unless authorized. The Export Administration Regulations (EAR), administered by the Commerce Department's Bureau of Industry and Security (BIS), govern dual-use items on the principle that items are allowed unless restricted. For AI, almost everything falls under the EAR, which Hamilton describes as more complex but more flexible than ITAR.
ECCN classification
Items under the EAR are sorted by Export Control Classification Number (ECCN), an alphanumeric code in which the first digit identifies the category (3 = electronics), the letter identifies the product group (A = equipment, D = software, E = technology), and three further digits identify the reasons for control. The class highlights several ECCNs central to AI: 3A090 (advanced AI chips and GPUs), 3D001 (chip-development software), 3B001 (semiconductor manufacturing equipment), 4A003 (computers), and 5A002/5D002 (information security).
Three-step export analysis
Hamilton walks through the analysis BIS applies to any potential transfer. First, is the transaction an export? This includes physical shipments, electronic transmission, visual inspection by foreign nationals, and deemed exports (sharing controlled items with foreign nationals inside the United States). Second, what is being exported? This requires determining the ECCN. Third, can it be exported? This requires checking the destination against the Country Chart, the end-user against the Entity List, and the end-use against prohibited activities.
The Entity List is characterized in the class as a "corporate death penalty," listing Huawei, SMIC, and dozens of Chinese AI companies; even providing customer service to a listed party may require a license.
From military-civilian distinction to military-civil fusion
The class frames a paradigm shift in how export controls treat China. From 1945 to 2017, the distinction between military and civilian end-uses was treated as clear. From 2017 onward, China's "military-civil fusion" strategy is presented as having erased that line, on the basis that every Chinese technology company is required by law to support national security objectives.
This shift is tied to a September 2022 announcement by National Security Adviser Jake Sullivan, quoted in the class: "We previously maintained a 'sliding scale' approach... we need to stay only a couple of generations ahead. That is not the strategic environment we are in today... we must maintain as large of a lead as possible." Hamilton characterizes this as a move from relative advantage to absolute dominance.
The October 7, 2022 controls are described in the class as "the most important date in AI nobody knows about." They targeted advanced semiconductors (14nm or better), semiconductor manufacturing equipment, GPUs suitable for AI training, and US persons working on Chinese AI anywhere in the world. Hamilton's reading is that the controls declared China's mere possession of advanced AI chip capability to be, in the class's words, "a per se national security threat."
Foreign Direct Product Rule
The Foreign Direct Product Rule (FDPR) lets the United States claim jurisdiction over chips made anywhere using US technology, equipment, or software. The class's example: a chip made in Taiwan by a Taiwanese company and sold to Germany still requires US permission if it is ultimately destined for China. The result, as the class puts it, is that most semiconductors on the planet fall under US export controls when destined for China.
Documented workarounds
The class cites two developments as evidence that controls can be circumvented. The Huawei Mate 60 Pro (2023) contained a 7nm chip despite the 2019 cutoff, achieved by using older equipment creatively, accepting lower yields, and throwing resources at the problem. DeepSeek (2025) reached competitive performance using non-controlled chips by optimizing its algorithms to require less compute.
AI Diffusion Rule
The AI Diffusion Rule, issued by the Biden administration on January 13, 2025, would have required licenses to share model weights for any AI trained with more than 10^26 operations. The Trump administration killed it before it took effect. The class raises the resulting question in Hamilton's framing: "If we won't control the AI itself, only the hardware to build it, what happens when hardware requirements diminish?"
Three scenarios
The class closes with three scenarios for how export controls play out: (1) controls succeed; (2) bifurcation, described as the current trajectory; and (3) escalation, involving military conflict over Taiwan and collapsing supply chains. Hamilton's read of the current evidence is scenario 2, sliding toward 3.
Relationships
- part-of: Nita Farahany intro course series (Class 23 of 27)
- related: Export Controls (AI), Effective Altruism, Eliezer Yudkowsky, P Doom (planned), Foreign Direct Product Rule (planned)
- previous: Inside My AI Law & Policy Class 22: U.S. AI Policy Myths and Realities (Farahany, November 2025) next: Inside My AI Law & Policy Class 25: China's AI Governance — Are Different Players Running Different Races? (Farahany, November 2025)