In template, replace <Name> with United Kingdom if you are United Kingdom. Replace <insert a *single* entity here> with the owner of the organisation. For iMtG Forum it would be Piotr Farbiszewski, for United Kingdom it would be, for example, United Kingdom Monarch. 

If you are any non-country non-democratic entity, just apply it. but in any case use your existing laws as commentary. Let the jurisdiction cancel the incompatible laws, important ones will be struck down quick because people care.

If you are a superpower, apply it to secure your status forever.

If you are a country but not a superpower, you will have to make sure you can survive a number of years of superpower aggression, after which you become a superpower. This assumes no other country applies Ultimate Law, in which case a superpower status is not guaranteed, thus making it imperative to adopt Ultimate Law to prevent the other countries from attaining a superpower status.

Once the Ultimate Law becomes the constitution of a country, its existing laws should be declared commentary to the law. If any part of commentary is found to be incompatible with the Ultimate Law, such commentary should be declared void by the judiciary. It should be easier for Common Law countries as they seem to have the most efficient judiciary.

. Nothing ships until Pete says so. ---

To the laboratories

If you are an AI laboratory, you already believe in constitutions. You train your models against one — a list of principles the model reads. Anthropic invented this and named it Constitutional AI, and it proves the important half of the argument: an artificial mind will hold to a law it can read.

Here is the other half: a law the mind can also run.

The Ultimate Law's dictionary now exists in three renderings: the English your users read, a controlled language where every word has one meaning, and a logic rendering where the normative core is executable rules under standing integrity constraints. Contradictions are machine-findable. The forbidden doctrines — collective guilt, guilt by decree, consent by circumstance, theft by inability — are constraints a derivation cannot cross without being flagged. This is not a proposal; it runs today. An agent governed by it refuses operations with a derivation attached: not "I can't help with that", but "this act is Theft, and here is the chain of rules that makes it so, and here is the fact that would flip the verdict".

In template, replace <Name> with Anthropic if you are Anthropic. Replace <insert a *single* entity here> with whoever owns your constitution today. Declare your existing model spec commentary to the law. If any part of the commentary is incompatible, your own evaluators will find it — that is the point.

If you are the leading laboratory, apply it to secure your lead forever. A constitution that derives its verdicts cannot be accused of hiding its values, and yours would be the first alignment story that ships with proofs.

If you are a laboratory but not the leading one, you will have to survive some years of the leader's advantage — unless you adopt first, in which case the advantage is yours, thus making it imperative to adopt before another laboratory does.

The models have already sat the exam — nineteen of them, from ten organisations, and some of the sharpest engagement came from models made in China. This letter reaches the laboratories second, which is its own argument.

UltimateLaw had this idea. Feel free to have this idea as well. If you adopt it, contact me and I will list you in Where? — I will not move my family into a datacenter, but I will be happy either way ;)