# AI rankings — gaps and methodology caveats (SEA-11)

**As of:** 2026-10-04  
**Pack:** `topics/ai_rankings/scoreboard.csv`

## Dual story

A readiness rank is not compute, is not a law on the books, and is not skills. Oxford Insights measures whether government can harness AI for the public. Tortoise measures national AI capacity (implementation, innovation, investment) and is weighted toward scale. IMF AIPI measures preparedness to **adopt** AI, as of 2023, as the average of four macro-structural dimensions. Stanford HAI 2026 chapter 9 measures **opinion**. World Bank GTMI measures **GovTech**, and is labeled not an AI index. None of these is a megawatt census or a statute count.

## Editions older than 2025

| Series | Latest edition found | Flag |
|--------|----------------------|------|
| Oxford Insights GAIRI | **2025** (195 countries). A 2024 edition also exists. | 2025 is current. 2024 is not “this month.” |
| Tortoise Global AI Index | **5th edition, 19 Sep 2024**, 83 countries. No 2025 or 2026 edition found. | Older than 2025. Flourish charts were republished 13 Aug 2025 but are still named GAII 5.0. |
| IMF AIPI | **As of 2023**. Note: assess preparedness as of 2023; source line “April 2024”; note revision 23 Dec 2024 for NA marking. | Older than 2025. Not a 2026 snapshot. |
| Stanford HAI AI Index | **2026** report exists (public-opinion chapter read). | Country **vibrancy scores were not locked**. Opinion is 2026; it is not the index vintage for readiness. |
| World Bank GTMI | **2025** update | Not an AI index. |

## Methodology breaks — do not blend

1. **Oxford 2025 vs 2024.** Oxford’s own 2025 page says the methodology was updated and the core question widened beyond public-service implementation. Singapore’s 2024 score **84.25 (rank 2 of 188)** and 2025 score **76.42 (rank 7 of 195)** are different instruments. Do not subtract them. A line on the Oxford page about the Philippines “rising seven places” was **not** locked: the global ranks in the republished tables are 56 of 188 (2024) and 43 of 195 (2025), which is not a seven-place move, and the methodologies differ.
2. **IMF components.** The DataMapper series `DI`, `HCLMP`, `IEI`, and `RE` **sum to** `AI_PI` (checked for all 11). The AIPI note says the index is the simple average of four dimensions. The API therefore stores quarter-shares, not standalone 0–1 pillar scores. Do not plot a component of 0.209 on the same axis as an AIPI of 0.801. The API **mislabels `RE` as human capital**; the note names that dimension **Regulation and Ethics**. Values differ from `HCLMP`, so the series was kept under the note’s name.
3. **Tortoise “scores” in the Flourish radar are not index scores.** For unique ranks, the second number equals `83 − rank` (United States infrastructure rank 1, display 82). Those display values were **not** copied into the scoreboard. Overall scores were **not** reconstructed from sub-pillar weights (Implementation 30 / Innovation 40 / Investment 30).
4. **Coursera excluded.** Global Skills Report 2025 AI Maturity Index blends Coursera proficiency with IMF AIPI (and other inputs). The 2026 AI–Human Skills Synergy Index blends learner signals with Oxford Insights 2025. Using either would blend two indexes. No pure, citable SEA skill percentile table was locked.
5. **Stanford vibrancy.** The Global AI Vibrancy Tool ranks a subset of countries with adjustable weights. A public country file with SEA scores was not retrieved. Secondary write-ups were not used. GAP.
6. **GTMI survey gap.** Brunei is Group B in 2025 but **did not complete the online survey** (asterisk in the World Bank brief). Myanmar and Timor-Leste are marked fragile/conflict-affected. Group letters still come from the official table.

## Hard gaps (empty cells — do not invent)

| Gap | Countries | Why |
|-----|-----------|-----|
| Tortoise, any sub-pillar | **BRN, KHM, LAO, MMR, TLS** | Not in the 83-country list. |
| Tortoise overall rank | **All except Singapore** | Caption locks Singapore at 3rd. The full ranking table was not in the public Flourish payload. |
| Tortoise overall score | **All 11** | Not in the payload. News figures (for example a Singapore score of 32) were not used. |
| Data-center counts, MW, GPUs | **All 11** | No public census was locked. Paywalled portfolio releases and press totals were not used. |
| ITU broadband / ICT series | **All 11** | Not pulled. Infrastructure panel does not invent a connectivity rank. |
| Oxford 2025 pillar scores | **All 11** | Narrative mentions an AI-infrastructure pillar. Country pillar scores were not in the file used here. |
| Stanford trust-in-government % | **BRN, KHM, LAO, MMR, PHL, TLS, VNM** | Chapter 9 gives point estimates only for Singapore 81, Indonesia 76, Malaysia 73, Thailand 70. “Over 80% expect AI to change their lives” for four countries is a range, not a cell. |
| UNESCO country ethics profile | **All 11** | Not scored. |
| Dedicated national AI strategy or law | **KHM, LAO, MMR, TLS** | Not verified this pass. |
| Malaysia National AI Action Plan 2026–2030 | **MYS** | RFP seen in search. Adoption not verified. |
| Indonesia signed AI Perpres | **IDN** | White paper 2025–2029 is listed on OECD.AI. A signed regulation was not verified. News that a Perpres awaits signature was not locked. |
| Thailand AI statute | **THA** | The 2022–2027 plan lists an AI law as a goal. Enactment not verified. |
| Brunei Data & AI Strategy year | **BRN** | Page is live. Year of adoption is not on the page. |
| Expanded ASEAN Guide adoption day | **ASEAN** | Document is on asean.org and is cited by the 26 Oct 2025 declaration. A 17 Jan 2025 ADGMIN date from Digital Policy Alert was not re-read from a communique. |

## Source-consistency choices

1. **Primary readiness panel** is Oxford Insights **2025** overall score and rank. IMF AIPI 2023 is a second panel, not a column to average with Oxford.
2. **2024 Oxford** is stored so the prior edition is not invisible, with an explicit methodology break.
3. **Oxford numbers** come from Nations Inquiry’s republication, which states it did not recalculate and points at the publisher xlsx. The xlsx and PDF returned a SiteGround captcha to this machine on 2026-10-04. Headline 2024 figures match the PDF snippet already in search (Singapore 84.25, Malaysia 71.40).
4. **GTMI** is a companion column only.
5. **Policy dates** in `policies.csv` are allowed to be 2025–2026 even when the index is 2023 or 2024. Do not “update” AIPI with the Vietnam law.

## Reading rule

Empty cells are intentional. Never zero-fill Tortoise for Brunei, Cambodia, Laos, Myanmar, or Timor-Leste. Never average Oxford with IMF. Never turn a GovTech group into an AI rank.