A legal witness to A ≤ E, and where it stops.
The Council of Europe / Alan Turing Institute primer is the cleanest citable statement of the paradigm the framework calls governance at rest: assess risk, set principles, impose obligations, audit for compliance. Authority is granted once, at design and procurement time, and never recomputed along the path.
Read carefully, it is also something more interesting — an independent witness. Working from human-rights law with no physics, it arrives partway onto the framework's own zeroth law, and then stops at a boundary that can be named exactly. What follows is that walk.
A 46-page plain-language companion to the CAHAI Feasibility Study (adopted December 2020), the document that fed the CAHAI → CAI line and the Council of Europe Framework Convention on Artificial Intelligence (opened for signature September 2024, the first binding international AI treaty). The primer is not a research contribution; it is a bridge document, written to widen stakeholder consultation. Judge it as pedagogy and policy scaffolding, not as new theory.
Citation caveat. The primer's front matter describes its license as “Creative Commons Attribution License 4.0 which permits unrestricted use,” but links to CC BY-NC-SA 4.0 (NonCommercial-ShareAlike) — a different, more restrictive license that is neither “unrestricted” nor plain attribution. If you reuse the work, the linked license governs, not the prose.
A single crop-yield example carried through Design → Development → Deployment lets human choice enter visibly at every stage, rather than being asserted. This is the strongest pedagogy in the document and the reason it works as onboarding.
Chapter 5 pins each of nine principles to substantive rights with real ECHR anchors (Arts. 2, 5, 6, 8, 10/11, 13) and to concrete Member-State obligations. This is where the primer stops being a glossary and converts values into justiciable hooks.
It names soft law's failure mode plainly — “tokenistic and largely rhetorical” — and argues for combining voluntary and mandatory governance rather than overselling principles.
Claim under test. The primer's precautionary, oversight-tailored governance is the same operation as A ≤ E, in a different substrate — law rather than physics.
Level-3 homology (direction-preserving, not a full isomorphism)
The primer's precautionary line holds that where uncertainty about a system's risk rises, oversight must rise and, at the limit, use must be prohibited. That preserves A ≤ E's direction exactly: as the environment's capacity to bound a system falls, permitted autonomy must fall, reaching zero at the limit. You can perform the action in one substrate — drop enforcement certainty — and predict the result in the other — drop permitted autonomy.
The primer reaches this from human-rights law and the precautionary tradition, with no physics anywhere in the argument. That makes it a juridical witness converging on A ≤ E from a starting position none of the framework's other witnesses — the bee, the acequia, Ostrom, the neural, the economic — occupy.
It is not bijective and not continuous. The primer determines the bound once, coarsely, at design and procurement time. A ≤ E recomputes it continuously along the path. Genuine homology, honestly downgraded from isomorphism.
The primer's frame holds where that ratio is at or below 1: episodic decisions, a human institution in the loop, accountability at audit and procurement timescales. In that regime, approve-once governance is adequate and the primer is right.
It breaks where the ratio is far above 1: agents acting at machine tempo, composing and federating, with no human carrier to hold the obligation between assessment points. There, approve-once plus periodic audit cannot recompute authority fast enough, and trajectory-native authority — A ≤ E recomputed along the path — is required. That interval is exactly what KTP governs and the primer cannot.
There is no trajectory-as-identity primitive anywhere in the document. Without it, authority cannot be recomputed along the path, so the frame is governance at rest by construction, not by oversight. The primer's own “dynamic assessment” and “continuous monitoring” (Chapter 7) are compliance bolt-ons on a static skeleton, not an authority substrate.
“You found KTP and now a fifty-year-old precautionary principle “secretly” contains A ≤ E. Hammer, meet nail.”
The claim is only a bounded, directional correspondence, and it comes with its regime boundary attached — tempo. A projected isomorphism claims total mapping; a genuine one names where it breaks. The boundedness is the evidence, not a hedge against it.
The source
Leslie, D., Burr, C., Aitken, M., Cowls, J., Katell, M. & Briggs, M. (2021). Artificial Intelligence, Human Rights, Democracy, and the Rule of Law: A Primer. The Council of Europe.
read the primer ↗