AI-SAF

AI Strategic Autonomy Framework

If a supplier or a government cut off your AI tomorrow, what still works? A stress test for AI dependence — free, open, runs in your browser.

Not affiliated with any AI Safety Institute.

Two axes

Think of it as a stress test. Two readings, never averaged:

  • Resilience

    60 points

    “What still works if a supplier cuts us off tomorrow?”

    23 indicators · 9/9 pillars

  • Autonomy

    40 points

    “Who decides our technology path?”

    15 indicators · 6/9 pillars

Both readings come from the same 38 answers, divided between the two axes with no overlap. The report keeps them apart; they are never averaged into one number.

Two ways in

How it works

  1. 01

    Nine pillars

    Data, models, infrastructure, hardware, manufacturing, software, human capital, regulatory, culture — weighted by how much each constrains a firm.

  2. 02

    38 indicators

    Each answered against written anchors, with a confidence rating — verified, estimated or guessed — per pillar.

  3. 03

    One score, two axes

    A 0–100 score with an uncertainty band, plus separate resilience and autonomy readings. They diverge, and that is the point. The result is a finding, not a grade.

Pillar weights, drawn to scale

sample profile · 61 / 100, ±4

  • Models9.0 / 16
  • Data10.0 / 14
  • Infrastructure and energy6.0 / 14
  • Human capital9.0 / 13
  • Regulatory (compliance)10.0 / 12
  • Hardware architecture3.0 / 10
  • Software stack8.0 / 10
  • Culture and operating model6.0 / 10
  • Manufacturing0.0 / 1
Track lengths are proportional to pillar weight, scaled so the heaviest pillar (Models, 16) spans the full width. The fill is the share earned — Manufacturing really is worth 1 point.

From the record

27 July / 2 August 2026 · Regulatory instruments

The Digital Omnibus on AI (proposed 19 November 2025) entered into force on 27 July 2026, deferring the AI Act’s high-risk obligations: stand-alone Annex III systems to 2 December 2027, and AI embedded in Annex I regulated products to 2 August 2028. The GPAI provisions were not deferred: on 2 August 2026, Article 50 transparency obligations, Commission penalty powers over GPAI providers, and full national market-surveillance authority took effect as originally scheduled.

Gibson Dunn, “EU AI Act Omnibus Agreement — Postponed High-Risk Deadlines and Other Key Changes”

Primary source ↗

See 36 more in the Evidence Base →

What this is not

Not a ranking.
It does not say who is best at AI. It says what survives a cut-off.
Not a maturity model.
A low score is a finding, not a failure grade.
Not a vendor product.
No account, no email, no sales call. The instrument is the deliverable.