Contents
Taiwan Strait, AI and conflict: a discussion guide
A first proposal from The Unjournal team for the working group's scope, outlining four focus areas with main questions and a handful of readings for each. A starting point for discussion.
Comment directly on this page: select any passage and choose Annotate, using Hypothes.is. How it works
AI-assisted draft · How this page was made
How this page was made: sources, models and what a person has checked
| Written by | Originally drafted by Claude Opus 5.5 (Anthropic), then revised by Codex (OpenAI) following Unjournal team comments, 28 October 2026. |
|---|---|
| Scope and framing | The working group's scope note (26 October) and comments from The Unjournal team. The four areas are a proposal for the full group to review. |
| Overviews and readings | Overviews and reading notes drafted with AI and revised by The Unjournal team over several rounds, drawing on the convenor's annotated bibliography and the longlist (258 items found by AI research agents with web search, 23 and 28 October; see that page's "How this list was made"). |
| Checks and limits | Unjournal team comments have informed the scope and structure. The two arXiv outlets and the South Africa export-control correction were checked against source pages. Reading summaries have not been systematically checked. |
| Corrections | Suggest an edit or addition, comment with Hypothes.is, or email contact@unjournal.org. |
What the group is asking
The aim is a short list of pivotal questions: open questions about the world where better evidence could improve a decision. A pivotal question meets three criteria:
- A decision depends on it. An identifiable actor (a government, AI lab, funder or advocacy organisation) would plausibly act differently depending on the answer.
- We are genuinely uncertain. Informed people disagree or hold a wide range of views, so better evidence could shift them.
- Research can make credible progress. A definitive answer isn't required. Narrowing a range, testing a key assumption or identifying which scenario we're in all count.
Defining the scope
The four areas suggested below are a first-pass sketch from The Unjournal team, for the full working group to review and discuss. Areas 1–3 focus on how AI development, expectations about its strategic value, and competition over chips and compute affect the risk of conflict in the Taiwan Strait.
We propose leaving military applications of AI outside this scope.
Where the line falls
To keep the focus manageable, questions about the military applications of AI (its use in weapons, operations, command and control, intelligence or crisis decision-making) and their effects on conflict dynamics would be outside the scope.
That separation isn't perfectly clean. Part of why states value AI is its anticipated military payoff, which feeds race and preventive-war incentives. So beliefs about AI's future strategic value, including its military value, are treated as an input, without studying how AI actually performs in military use. The group would ask "what happens when Beijing believes Washington is about to gain a decisive AI advantage?", not "would AI-enabled ISR actually find China's submarines?"
- In scope: AI as the stakes. AI development shapes conflict risk because states expect it to shift relative power, compete to control its inputs, use dependence on it as leverage, and negotiate over it.
- Out of scope: AI as an instrument. How AI is used in weapons, military operations, command and control, intelligence, or crisis decision-support.
One open question is whether to also focus on catastrophic or existential risk from advanced AI itself (area 4).
Four possible areas (first-pass sketch) AI-assisted, team-revised
Four proposed focus areas, each with a short overview, a main question and starting readings. The group can refine these questions, suggest other publications or propose a missing area.
Suggest an edit or addition
What should we change in these four areas? Suggest edits, missing publications, missing questions, or a whole area the group should consider. For a proposed question, which decision could better evidence inform, and for whom?
Suggestions and contact details go privately to The Unjournal team.
Background readings
Estimates of how likely a war or blockade is, and what it would cost, provide context for the four areas. A few starting points:
Out of scope. Work on military uses of AI is on the longlist under "Outside scope: AI as an instrument", in case the group revisits the boundary. Agreements and norms about military AI (such as the REAIM process and the 2024 statement on human control of nuclear launch decisions) are kept in scope as evidence about cooperation; the group may want to check that line.
Comment on any passage with Hypothes.is
This page has Hypothes.is switched on, so you can attach a comment to any sentence. It is the easiest way to say "this summary is wrong" or "add this source" exactly where it applies.
- Select some text on the page. A small Annotate button appears; click it.
- The first time, sign in or create a free Hypothes.is account (it asks for a username and email).
- Write your comment and choose Post to Public.
To read other people's comments, open the sidebar with the arrow tab at the top right of the page; highlighted passages have comments attached.
Note: public annotations are visible to anyone who opens this page with Hypothes.is, under your Hypothes.is username. If you would rather not be publicly linked to a view on Taiwan, use the suggestion box (sent privately to The Unjournal team) or email contact@unjournal.org.