## The narrowed scope

The working group asks what research would most help stakeholders prevent war in the Taiwan Strait. It focuses on **how advanced AI development changes strategic incentives and relationship patterns between the US, China, Taiwan and their partners.**

- **In scope: AI as the stakes.** AI development shapes conflict risk because states expect it to shift relative power, compete to control its inputs (chips, fabs, compute), use dependence on it as leverage, and negotiate over it. *Beliefs* about AI's future strategic value, including its military value, are in scope as an input: "what happens when Beijing believes Washington is about to gain a decisive AI advantage?"
- **Out of scope: AI as an instrument.** How AI is used in weapons, military operations, command and control, intelligence and surveillance, crisis decision-support, wargaming with language models, or AI-enabled disinformation and cognitive warfare.

The outcome of interest is open for the group to decide: preventing war in the Strait, and possibly also how war risk interacts with catastrophic risk from advanced AI itself (trade-offs or complementarities between keeping the peace and making AI safe, and whether US–China cooperation on AI governance and safety is possible amid Strait tension). Work on that interaction is in scope.

## Categories

Give every item exactly one `category`:

| ID | Category | Covers |
|---|---|---|
| `race` | AI race dynamics and strategic incentives | Expected AI-driven power shifts and US–China–Taiwan incentives: preventive war, commitment problems, windows of vulnerability, race narratives and their critics, beliefs about decisive advantage, compute advantage as a trigger |
| `shield` | The silicon shield and chip dependence | Whether Taiwan's chip centrality deters attack, invites coercion or becomes a target; how onshoring and diversification change that |
| `taiwan-actor` | Taiwan's leverage and choices through its supply-chain position | Taiwan's tech diplomacy, alignment with export controls, economic statecraft by or against Taiwan, Taiwan's place in AI governance |
| `cooperation` | US–China AI governance, cooperation and the safety link | Feasibility of US–China cooperation on AI risk amid Strait tension; how war risk and AI catastrophic risk interact; export controls and compute governance as they affect strategic incentives |
| `background` | Background: conflict risk, costs and forecasting | Likelihood and costs of a Strait conflict or blockade, interdependence, forecasting and early-warning methods: useful context, but not about AI as the stakes |
| `out-instrument` | Outside scope: AI as an instrument | Military uses of AI and their effects on conflict dynamics, AI-enabled information operations, LLM wargame behaviour |
| `out-other` | Outside scope: other | Not about the Strait, AI geopolitics or conflict risk |

Also give `scope_status`: `core` (race, shield, taiwan-actor, cooperation), `background`, or `out`; and a one-sentence `scope_note` saying why.

## Search rules for the follow-up

- Start from the convenor's annotated bibliography (listed in your brief): verify each item, find a stable link, and add the ones not already on the list. Mark them `"convenor_bibliography": true`.
- Then search for more work in the four core categories. Search in English, Traditional Chinese (矽盾, 晶片 地緣政治, 科技外交, 出口管制 台灣), Simplified Chinese (硅盾, 中美 人工智能 竞争, 人工智能 治理 合作, 预防性战争) and Japanese (シリコンシールド, 半導体 台湾 地政学).
- Expert analytical commentary in policy outlets (for example Lawfare, The Diplomat, War on the Rocks, Lowy Interpreter) now qualifies when it makes a distinct argument people in this space cite; give it `type: "analysis_commentary"`. News reports still do not.
- Theses and dissertations qualify; give them `type: "thesis"`.
- Do not add items in `out-instrument`.