# 来源清单 / Source register

核验 / Reviewed: 2026-09-08

## M1 · Metix AI

[OpenAI, Anthropic and xAI Talent Structure Benchmark](https://metix.ai/reports/mapping/openai-anthropic-xai-talent-structure)

发表: 2026-07-23; 修订: 2026-08-20

职能及前雇主：美国在职档案，OpenAI n=5,313，Anthropic n=2,612；未覆盖 Google DeepMind。

商业人才数据库样本，可靠性有限：精确快照、分类优先级、MTS 归类规则、合同工边界及独立质量审计未披露。只转录原报告标签，不视为真实编制。

Functions and prior employers: US current-employment profiles, OpenAI n=5,313; Anthropic n=2,612. Google DeepMind is outside scope.

Commercial talent-database sample, with limited reliability: exact snapshot, classification precedence, MTS assignment, contractor boundary and independent validation are not disclosed. Labels are transcribed, not treated as actual staffing.

## M2 · Metix AI

[Frontier AI Labs Talent Landscape & Flow Map](https://metix.ai/reports/mapping/frontier-ai-labs-talent-2026)

发表: 2026-06-11; 修订: 2026-08-20

约 2026 年上半年；美、英、法。三家技术池分别 n=2,628 / 1,251 / 2,837；技术池排除 solutions/product/GTM。

按 title/headline 识别，未独立验证。Google 主体名下的 DeepMind 人员可能漏计；技术池层级示例与部分排除项存在边界张力。原文留存排序叙事与表值不一致，本页不采纳该叙事。

Approximately H1 2026; US, UK and France. Technical pools n=2,628 / 1,251 / 2,837; solutions/product/GTM excluded.

Title/headline classification without independent validation. DeepMind staff listed under Google may be missed; some hierarchy examples blur stated exclusions. The retention narrative conflicts with its table and is excluded here.

## B1 · BCG Henderson Institute

[Where Will Tomorrow’s AI Geniuses Go?](https://web-assets.bcg.com/4c/15/9ce4a0274a7cafc0d9a9b13ae32c/rapport-bcg-where-will-tomorrows-ai-geniuses-go-july2025.pdf)

发表: 2025-07-21

2024 年 H-1B 相关职位工资；PDF 第 7 页说明、第 8 页 Exhibit 5.2。

图注列 DOL、Levels.fyi、BCG 分析。未披露各公司记录数、均值／中位数、职级构成和法律实体；不含奖金与股权。招聘者调查 N=40 不是薪资样本量。

2024 H-1B-related position salaries; explanation on PDF page 7 and Exhibit 5.2 on page 8.

Chart cites DOL, Levels.fyi and BCG analysis. Company record counts, mean/median, seniority mix and legal entities are not specified. Bonuses and equity excluded. The recruiter survey N=40 is not the salary sample.

## S1 · SignalFire

[State of Talent Report 2026](https://www.signalfire.com/blog/signalfire-state-of-talent-report-2026)

发表: 2026-06-22

第 5 节 Frontier AI Labs：OpenAI、Anthropic、Google DeepMind、xAI、Mistral、Cohere 六家汇总。

公开版本未给这项分析的逐公司职能表及样本量。六家合计不能拆给本篇三家。

Section 5 Frontier AI Labs: an aggregate of OpenAI, Anthropic, Google DeepMind, xAI, Mistral and Cohere.

Public version does not provide company-level function tables or sample counts for this analysis. A six-lab aggregate cannot be allocated to the three companies.

## G1 · Google DeepMind

[Announcing Google DeepMind](https://deepmind.google/blog/announcing-google-deepmind/)

发表: 2023-04-20

官方组织边界：DeepMind 与 Google Research 的 Brain 团队合并为 Google DeepMind。

组织说明，不是职能人数披露；不能用 Google 或 Alphabet 全员替代。

Official organizational boundary: DeepMind and Google Research’s Brain team combined into Google DeepMind.

Organizational announcement, not a staffing breakdown. Google or Alphabet headcount cannot substitute.

## G2 · DeepMind Technologies Limited / Companies House

[Report and Financial Statements, year ended 31 December 2024](https://find-and-update.company-information.service.gov.uk/company/07386350/filing-history/MzQ4Mjk4NjY3OWFkaXF6a2N4/document?format=pdf&download=0)

发表: 2025-10-02 (filed)

英国法律实体 07386350；印刷第 3 页 Our people，PDF 第 4 页；第 20 页附注 7。

明确人员由其他 Alphabet 集团公司直接雇用，且成本包含集团转收。本实体财报不能当作 Google DeepMind 全球员工名册或直接计算全员平均薪资。

UK legal entity 07386350; printed page 3, Our people (PDF page 4), and note 7 on printed page 20.

Workforce is directly employed by other Alphabet group companies and costs include group recharges. These accounts are not a global Google DeepMind roster or a basis for its average employee pay.

## A1 · Anthropic

[Reflections on our Responsible Scaling Policy](https://www.anthropic.com/news/reflections-on-our-responsible-scaling-policy)

发表: 2024-05-20

官方披露当时约 8% 全员从事 security-adjacent areas。

历史时点、交叉类别；没有人数及总分母人数。Security-adjacent 不等于 AI safety、法务或政策团队，不能与它们相加。

Official disclosure that around 8% of employees worked in security-adjacent areas at that time.

Historical, cross-functional category. Employee counts not supplied. Security-adjacent does not mean AI safety, legal or policy, and must not be added to those categories.

## X1 · Metix AI

[OpenAI Engineering Talent X-Ray](https://metix.ai/reports/mapping/openai-engineering-xray-2026)

发表: 2026-06-15

约 2026 年上半年，OpenAI 技术档案 n=3,041。

与 M2 地域及分类口径不同。职业年限从最早记录起算；历史雇主可重复。当前页引用 Anthropic 的旧博士比例，本页不采纳该转述。

Approximately H1 2026, OpenAI technical profiles n=3,041.

Different geography and classification from M2. Career duration starts at the earliest record; historical employers overlap. A stale Anthropic PhD figure in the current page is not adopted.

## X2 · Metix AI

[Anthropic Engineering Talent X-Ray](https://metix.ai/reports/mapping/anthropic-engineering-xray-2026)

发表: 2026-06-15

当前原文技术档案 n=1,281；约 2026 年上半年。

旧索引样本与当前原文不同。年限有效样本数未单列；历史雇主每人每公司一次，多公司可重复；排除大学，不可推断学术背景缺失。

Current original page: technical profiles n=1,281; approximately H1 2026.

Older indexed sample differs from the current source. Valid career-duration n is not separately given. Historical employers count once per person/company and overlap across companies; universities excluded.

## R1 · Revelio Labs

[DeepMind employee count and department mix](https://www.reveliolabs.com/companies/deepmind/employees)

发表: 2026-08

已读取 2026 年 3 月公司页；实体写作 DeepMind Technologies Ltd.。

未采纳：公开部门总计与总人数不一致，且同比增量与百分比不一致；全球团队实体边界未解明。不能用来补齐第三列。

March 2026 company data page reviewed; entity named DeepMind Technologies Ltd.

Excluded: department totals differ from headcount, and year-over-year absolute changes conflict with percentages. Global team boundary unresolved; not used to fill the third column.

## Z1 · Zeki Data

[Bellwethers of AI: OpenAI, Anthropic, Mistral AI](https://zekidata.com/report/bellwethers-of-ai-openai-anthropic-mistral-ai/)

发表: Not established / 未确认

已读产品介绍；完整报告需要申请访问。

本次未取得全文，不能写成机构未披露；其列出的第三家是 Mistral AI。

Product description reviewed; full report requires access.

Full report not obtained in this review; this is not a claim of non-disclosure. The third named company is Mistral AI.