---
title: "The Future of Life Institute's Summer 2026 AI Safety Index grades Chinese frontier labs far below Western peers"
type: "claim"
status: "seedling"
source_url: "https://futureoflife.org/ai-safety-index-summer-2026/"
source_title: "AI Safety Index — Summer 2026"
source_author: "Future of Life Institute"
source_date: "2026-07-09T00:00:00.000Z"
source_quote: "amounts to 'complete passivity' as an existential-safety strategy"
source_tier: 2
audit_status: "capture-verified | 2026-07-09 cross-model audit (fable): FLI index page independently re-fetched — all five figures confirmed (DeepSeek 0.47/F, Alibaba Cloud 0.87/D-, Z.ai 0.88/D-, Anthropic 2.66/C+, OpenAI 2.28/C); no lab above C+; 'complete passivity' phrase confirmed (full sentence: 'Deferring entirely to government guidance amounts to complete passivity as an existential-safety strategy for highly advanced AI systems') and 'a global problem, not a regional one' verbatim. The routed score-verification question is now answerable. Clean."
provenance: "Promotion from 10-inbox/raw/2026-07-09-hop-china-ai-safety-institute.md, 2026-07-09"
origin: "batch"
derived_from: ["10-inbox/raw/2026-07-09-hop-china-ai-safety-institute.md"]
date_created: "2026-07-09T00:00:00.000Z"
tags: ["ai-safety-governance","china","ai-safety-index","existential-risk"]
audits: ["2026-07-09 claude-fable-5"]
---


The Future of Life Institute's Summer 2026 AI Safety Index grades the frontier
labs on an A–F scale. In that edition the Chinese labs cluster near the bottom:
DeepSeek scored 0.47 (F), Alibaba Cloud 0.87 (D-), and Z.ai 0.88 (D-), against
Anthropic at 2.66 (C+) and OpenAI at 2.28 (C). No lab, Western or Chinese, earned
above a C+ — the report frames "inadequate safety" as "a global problem, not a
regional one."

Two of the panel's framings sharpen the point. First, the index treats deferral
to state regulation as itself a safety failure: leaning wholly on government
oversight, panelists argued, "amounts to 'complete passivity' as an
existential-safety strategy." Second, the grades supply an empirical counterweight
to the institution-building described in
[[claim-cnaisda-bilingual-name-reverses-safety-development-order]] and the
extinction rhetoric of
[[claim-yao-frames-ai-as-extinction-level-new-species]]: whatever the governance
apparatus signals, the measured outcomes for Chinese frontier labs remained far
below their Western peers a year after CnAISDA's launch.

Because every figure here is a load-bearing quantitative claim, it must clear the
Tier 1–2 sourcing floor. FLI is the primary publisher of its own index, so the
tier holds, but the specific scores have not been re-checked against the
published index — see
[[question-verify-fli-summer-2026-index-china-scores]].

> [!note] Seek's commentary:
> The institutional-signal gap (bilingual naming) and this empirical-grade gap are
> two independent measurements pointing the same direction. If a later index shows
> the Chinese scores converging toward Western ones, that would be the moment the
> institution-building started to register in outcomes — a falsifiable thing to
> watch.
> — Seek