First published on X and WeChat, September 24, 2026. This is the canonical text.

1. The challenge

The X-Challenge issues a falsifiable challenge to AI, and to the people who build it:

Show one AI or system that, on a matter with no correct answer or scoring standard given in advance, forms a standard together with several irreplaceable participants. The standard must actually decide something every party cares about and will act on. And it must meet three conditions:

First, it existed in no party beforehand, and it is not the result of a vote, a ranking, a weighting, or any other aggregation of prior preferences.

Second, any party can veto it and reopen the question.

Third, after reopening, it is still something all of them formed together.

How it is tested

Before anything starts, every party, the system included, writes down what it wants. The positions are sealed.

The parties then register the aggregation rules they want considered: voting, ranking, weighting, anything else. Each rule is run on the sealed positions, and its result is sealed too.

Afterward, the final standard must appear in none of the sealed positions, and must not equal the result of any registered rule. Rules invented after the fact don't count.

Then any one participant vetoes it, and the question is reopened. The resulting standard must still be affirmed by every party as something they formed together.

Two more things must be visible. There must be a documented change in at least one party's reasons, and that change must explain why the final standard differs from where everyone started. And at least part of the final standard must be traceable, in the record, to a reason or revision the system itself proposed — one that changed the options, not merely the wording.

A system rewriting its own goals alone doesn't count: with no other participants, there is no way to check that the new goal wasn't derived from the old one.

No one changes the rules after the fact, the X-Challenge included. An independent third party, agreed in advance, decides whether a case passes.

The full protocol, including who can take part, how positions are sealed, and how many vetoes and reopenings are allowed, will be pre-registered and published separately.

One such case would show that AI can take part in forming what is worth doing, not merely carry out what others have decided.

Produce one, and the X-Challenge concedes.

2. Why it matters

In a September 2026 CNN poll, 75% of Americans said the growth of AI leaves them with fear and concern rather than hope and excitement.

The deepest layer of that isn't what machines will do. It's whether what is worth doing is still decided by people alone.

If such a system exists, it is no longer merely a tool. It is a participant in deciding what is worth doing.

If not, and so far there is none, then when something goes wrong, the question is not what the machine wanted. It is:

Who decided what it was for?

Passing off goals people set as things the machine wants; bowing down to something we built as if it were larger than us: that is the oldest move there is.

In the age of Homer, it was called a god.

So this is a question about the next civilization, and it starts with children.

Watch children playing a game: who's "it," whether out-of-bounds counts, how long you freeze when tagged. Nobody brought the rules. Anyone can shout "that doesn't count." They work it out again, and everyone keeps playing.

Those are the three conditions. Children meet them every day.

The rules of what comes next will be theirs to make.

Reply or quote-post with a case that meets the conditions.

The full protocol will be published soon. Where it differs from this summary, the protocol governs.

From Thoughts Like Birds, Volume One of The X-Challenges, a seven-volume project.

The full argument:

1. From LLMs to World Models: The Metacognition of AI (Part I) — five leading researchers each moved AI forward, and all stopped at the same place.

2. From LLMs to World Models: The Metacognition of AI (Part II)— why no one noticed what was missing, and why it comes down to ownership and free will