Who decides what AI is for
Four governments, four Davos speakers and one standoff between the Pentagon and an AI lab: early 2026 showed there isn't one AI debate but several, each about something different.
Clients often ask us about cross-border AI compliance as if there were one set of rules to follow. There isn't. And the more I looked, the clearer it became that the disagreement isn't really about rules. It's about what AI is for, and who gets to decide.
Four governments, four answers
Rather than compare laws line by line, we looked at the United Kingdom, the European Union, the United States and China by their core ideas. Each writes different rules because each is protecting something different.
The UK sees AI as a growth engine. There's no dedicated AI law yet. Existing regulators apply five non-binding principles, and an AI Bill may arrive as early as late 2026. Until then, the absence of binding rules means legal uncertainty.
The EU treats AI as a product safety problem. The AI Act sorts systems by risk level and requires checks before they reach the market. High-risk systems in finance and insurance face strict documentation requirements and must keep a human in the decision.
The US sees AI as a way to stay ahead. The December 2025 Executive Order set out to cut federal regulation and stop states from writing their own AI rules. Some states are pressing ahead anyway.
China uses AI to strengthen national control. Instead of one broad law, it has a set of targeted rules, including interim measures on generative AI and labelling requirements for AI content. The updated Cybersecurity Law folds AI compliance into national security, and every algorithm must be registered with the government.
Growth, safety, competitiveness, control. Common ground between those four is hard to find, because they aren't answering the same question.
Four speakers at Davos
The same split showed up at Davos 2026. I went through the speeches of Alex Karp, Elon Musk, Dario Amodei and Yuval Noah Harari, not as an expert, just reading what they actually said. They weren't merely disagreeing. They were playing different games.
Karp (Palantir): AI is a weapon. He doesn't philosophise about humanity's future. He talks about victory and defeat, and expects AI to destroy jobs in the humanities.
Amodei (Anthropic): AI is a dangerous mind, "a country of geniuses in a data center." His fear isn't the enemy but losing control of the system itself.
Musk (Tesla, SpaceX, xAI): AI is a fire he keeps fanning. He warns about the Terminator while building robots. Someone will build godlike AI eventually, so he wants it to be him.
Harari (historian): AI is the end of humanism, "a knife that can decide to kill." He thinks in millennia and worries that humanity may lose its identity.
My bet on what happens next:
- Harari's philosophy will be applauded, then ignored.
- Musk's contradictions will be forgiven, because he delivers something tangible.
- Amodei's warnings will be heard selectively, when convenient.
- Karp's framework will shape actual policy, because it speaks the language of power.
When it stopped being theoretical
A few weeks later, the debate turned operational. US Defense Secretary Pete Hegseth reportedly gave Anthropic a 48-hour ultimatum: remove the safety guardrails for military use, or lose a $200M Pentagon contract and risk being blacklisted from the entire defence supply chain.
Amodei didn't budge. His two red lines: no fully autonomous weapons, and no mass surveillance of American citizens.
The Pentagon's argument was that once it buys a technology, how it's used is its call, the same way Boeing doesn't dictate how the military flies its jets. OpenAI, Google and xAI had already agreed to make their models available for any lawful military use.
Anthropic's argument was that AI isn't reliable enough to make life-and-death decisions without human oversight. As Amodei put it, the constitutional protections in military structures depend on there being humans who would disobey illegal orders, and fully autonomous weapons can't make that distinction.
Claude had reportedly already been used in the operation that captured Venezuela's Maduro. The question of where AI agents stop assisting and start deciding is no longer hypothetical.
Several debates, not one
Put the three together and the picture is consistent. We're not having one AI debate. We're having at least four, about labour, abundance, control and meaning, and different actors are answering different ones. Governments pick the question that matches what they want to protect. Leaders pick the one that matches what they're building.
For a business, that has a practical side. If you operate across borders, you aren't complying with "AI regulation"; you're complying with several philosophies at once, each likely to evolve in its own direction. Design for the strictest regime you face, keep a human in the decisions the EU would require one for even where nobody else asks yet, and watch where your vendors draw their lines. As the Pentagon standoff showed, the position your AI provider takes can become part of your own risk profile.
The technology is moving faster than governance. Whichever side you take, this is the conversation that will define not just AI's future, but ours.