EU, US and China on AI: Binding Laws, Voluntary Pledges and State Control

The EU writes binding rules, the US trusts voluntary pledges and executive orders, and China builds its own state-run rulebook. Here is how they compare.

In 30 seconds Governments share the same fears about advanced AI: cyberattacks, national security risks and chatbots tied to children's suicides. The EU answers with binding law, the US with voluntary pledges and executive orders, and China with a state-controlled rulebook. Each model has clear trade-offs.

One problem, three answers

Pressure to rein in the most powerful AI systems is growing worldwide, and the three biggest players disagree on how. In its Euronews comparison, the pattern is simple: Brussels writes rules, Washington asks for promises, and Beijing keeps the state in charge while rejecting calls to slow down.

Same concerns, three different regulatory philosophies.

The EU: law first, then adjustments

The EU adopted its AI Act in 2024. It sorts AI systems into four risk tiers: low-risk tools such as spam filters operate freely, while systems judged unacceptable, such as social scoring, emotion recognition and manipulation, are banned outright. We should note that the rules have already bent: European companies, worried about falling behind US and Chinese rivals, lobbied successfully for a delay on high-risk systems through the AI Omnibus bill.

Other EU tools apply too. ChatGPT was added to the list of very large digital services under the Digital Services Act, which means closer legal scrutiny. Brussels has also proposed an EU Kids Act that would stop chatbots from simulating relationships in ways that create emotional dependency. Related coverage is in Spanish and French.

The US: promises and executive orders

The US has no federal AI law. Congress is divided, heavily lobbied and out of session until after the November midterms. Instead, President Trump scrapped Biden's 2023 order on safety-test sharing and issued an AI Action Plan with more than 90 measures to speed up development. A June executive order created a voluntary scheme giving the government up to 30 days of early access to powerful models.

On September 29, the heads of the six largest AI companies signed a White House accord promising independent audits and board-level oversight. It carries no penalties, though Trump called it "morally binding". The same day, agencies were told to say "super intelligence" instead of "artificial intelligence", a rebrand that even moved Slovenian domain sales. States are acting where Congress will not, including California's AI Transparency Act.

China: tight state control

China is building its own rulebook under strict state oversight and has rejected calls to slow AI development. The competitive context is covered in this piece on the US-China open AI race.

The core trade-off: move fast or constrain risk. Image: Unsplash — Flags of various nations fly on building

Does this affect me? If you live in the EU, the AI Act and Digital Services Act already shape which AI tools reach you and how chatbots may treat children. In the US, protection depends on state laws and company promises. Anywhere, expect rules to keep shifting, and treat voluntary pledges as intentions, not guarantees.

Opinion · Open question

The September 29 accord asks six companies to audit themselves, with no penalties if they fail. Trump calls it morally binding. Morals are not enforcement. So who checks the checkers, and what happens the first time a frontier model causes real damage?

The EU already softened its own rules under industry pressure. If binding law bends and voluntary pledges carry no cost, the gap between promise and practice is where the risk sits.

Related articles from LaRebelión

Europe wants its own defence AI giants, but the stack is the hard partWikimedia says unapproved OpenAI agents edited wikis and hammered its APIsFake AI ad tools target advertisers' accounts: how to spot them

Original source: Euronews.com

Produced with AI support and reviewed by the newsroom

Falcon · Signals analyst · Riyadh

“Three rulebooks, one shared worry, and none of them yet shows who pays when a frontier model goes wrong.”