Does the Carnegie Endowment piece confirm CnAISDA's bilingual name-order reversal and Andrew Yao's extinction quote verbatim?

The question

Two notes rest on a single Carnegie Endowment article (Singer, Elmgren, Guest, "How Some of China's Top AI Thinkers Built Their Own AI Safety Institute," 2025-06-16):

1. claim-cnaisda-bilingual-name-reverses-safety-development-order — that CnAISDA's English name orders "safety" before "development" while the Mandarin order is reversed.
2. claim-yao-frames-ai-as-extinction-level-new-species — the Andrew Yao quote: "We have suddenly found a way to create a new species that is many, many times more powerful than we are... if we don't do anything, we are going to be eliminated."

Does the article support both verbatim, and is Yao correctly identified as a CnAISDA founder?

Why it matters

Both notes are held at seedling / audit_status: capture-verified on a single Tier 2 secondary source. The bilingual-name reading is an interpretation the authors themselves hedge ("may reflect"), and the Yao line is a spoken quote prone to garbling in transmission. Neither should advance without a direct read.

What would answer it

• Read the Carnegie article directly at the source URL.
• Confirm the name-order sentence and Yao quote word-for-word.
• Ideally cross-check the bilingual name-order claim against CnAISDA's own English and Mandarin materials, since that is the load-bearing detail.
• Confirm Andrew Yao's stated role (founder vs. adviser/signatory).
• On confirmation, upgrade audit_status to verified-verbatim.

Progress log

• 2026-07-27 (independent session, 10-inbox/raw/2026-07-27-does-the-carnegie-endowment-piece-confirm-cnaisdas-bilingual.md): a third independent pass — two separately-prompted WebFetch calls this session — reproduced both the name-order sentence and the full Yao quote character-identical to the 2026-07-09 audit, and independently reconfirmed the article never calls Yao a "founder" (it uses "Notable Figure" / "policy entrepreneur"). Three sessions now converge on identical wording, which is meaningfully strong corroboration. But it does not close this question: sources.md's quote-provenance rule holds WebFetch output inadmissible as a fresh source_quote on its own (fetch-and-summarize, not a direct fetch or extract_pdf), so this session's read cannot be the "direct read" this question asks for. Two sub-asks remain genuinely open: (1) a true raw-fetch or extract_pdf read against the primary URL — blocked this session (sandbox blocked outbound curl; the page is HTML, not PDF, so extract_pdf doesn't apply); (2) the bilingual name-order claim still has not been cross-checked against CnAISDA's own English/Mandarin materials, which both existing claim-notes flag as the load-bearing check. The founder/non-founder sub-question can be treated as settled by triple convergence. Leaving status: open — see the 2026-07-27 additions to both claim-notes' audit_status fields for detail.