Four AI Rulebooks Were Written in One Week. Most Countries Are Signing More Than One.

The week of July 14 produced four separate attempts to govern artificial intelligence, on three continents, with almost no reference to each other.

That is not gridlock. It is the actual shape of AI regulation now, and it changes what compliance means for anyone shipping a model.

## What landed

In Shanghai on July 17, Xi Jinping used his World AI Conference keynote to launch the World Artificial Intelligence Cooperation Organization, an intergovernmental body built around open-source norms and a call to stop treating AI as a national security matter. It drew the heaviest press coverage of anything we tracked that week. Indonesia signed the founding agreement the day before it was announced.

In Brussels, the European Commission ordered Google to open Android to rival AI assistants and share anonymised search data with competitors by January 2027. Not a framework. A binding decision with a date on it.

In Washington, reporting suggested the administration is weighing a FINRA-style independent regulator to vet frontier models. In the same stretch, OpenAI's policy chief argued that state bills in California, New York and Illinois are already converging into a de facto national standard, and that Washington should ratify it rather than fight it.

And a Rome Declaration tied AI directly to nuclear weapons policy, following a Vatican assembly of roughly 30 Nobel laureates.

## The part everyone is misreading

The instinct is to score this as a bloc contest. US rules versus Chinese rules, with the EU writing the fine print.

The dealmaking does not support that. In the same week WAICO launched, India and the EU agreed a joint roadmap on AI, semiconductors and 6G at their third Trade and Technology Council. Indonesia signed with Shanghai. Azerbaijan announced a regional AI centre serving Central Asia and the Middle East. These are not countries choosing a camp. They are countries joining every table that has a chair open.

We found the same pattern when we counted deals earlier this month: the US-versus-China frame kept hiding where Asian AI money actually moves. Governance is following the capital, and the capital is promiscuous.

The reason is simple. Nobody knows which rulebook will bind, so the cheapest strategy is optionality. Sign the treaty organization, keep the bilateral roadmap, wait to see which one has teeth in 2028.

## Why this is a cost, not a headline

Multiple non-exclusive regimes sound harmless. They are not, for the people building.

A model deployed across three of those jurisdictions now faces open-weight expectations from one, interoperability mandates from another, and possibly a licensing gate from a third. Greece is separately proposing prison terms for stripping watermarks off AI-generated media, which is a real criminal exposure attached to a technical detail most teams treat as metadata.

The compliance surface is not converging. It is multiplying, and each new venue adds obligations without cancelling the last ones. That is also visible in our federal AI regulation and regulatory scrutiny trend clusters, both of which have been climbing.

## What to do with this

**Stop tracking rulebooks. Start tracking enforcement dates.** Most of what landed this week is declaratory. One item is not: the EU gave Google a January 2027 deadline with a defined remedy. That is the only one currently capable of changing a product roadmap, and it is the template worth watching, because binding orders with dates are what other regulators copy when declarations stop working.

**Ask which regime your provenance layer answers to.** Watermarking and content provenance are becoming criminal-law questions in some jurisdictions, not trust-and-safety questions. If your answer to "who governs our watermarks" is a team rather than a statute, that gap is now worth pricing.

The bloc story makes for cleaner commentary. The optionality story is what the signatures actually show.